diff --git a/workspace/runs/ctt_base_context_obs_learned_dominance_basic_perdim_trainmax_train_to_test/command.txt b/workspace/runs/ctt_base_context_obs_learned_dominance_basic_perdim_trainmax_train_to_test/command.txt new file mode 100644 index 0000000000000000000000000000000000000000..3aac963f491c1b94431dae38365c6dfb0711ced6 --- /dev/null +++ b/workspace/runs/ctt_base_context_obs_learned_dominance_basic_perdim_trainmax_train_to_test/command.txt @@ -0,0 +1 @@ +python scripts/eval_learned_dominance_selector.py --calibration-input runs/ctt_base_context_obs_train_cal_perdim_trainmax_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_perdim_trainmax_rollout_comparison/combined_measured_candidates.json --eval-target-index data/cil_charts_rgb_refs/test/index.json --checkpoint-template runs/ctt_residual_base_context_obs_seed{seed}/model.pt --out-dir runs/ctt_base_context_obs_learned_dominance_basic_perdim_trainmax_train_to_test --feature-set basic --target utility_margin --k 8 --bootstrap-samples 1000 --no-markdown-report diff --git a/workspace/runs/ctt_base_context_obs_learned_dominance_basic_perdim_trainmax_train_to_test/config.yaml b/workspace/runs/ctt_base_context_obs_learned_dominance_basic_perdim_trainmax_train_to_test/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..51bac489d879b2c0579c264fe7f836e0a976a4f9 --- /dev/null +++ b/workspace/runs/ctt_base_context_obs_learned_dominance_basic_perdim_trainmax_train_to_test/config.yaml @@ -0,0 +1,18 @@ +bootstrap_samples: 1000 +calibration_input: runs/ctt_base_context_obs_train_cal_perdim_trainmax_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_perdim_trainmax_rollout_comparison/combined_measured_candidates.json +eval_target_index: data/cil_charts_rgb_refs/test/index.json +feature_set: basic +fit_objective: pointwise +k: 8 +no_markdown_report: True +out_dir: runs/ctt_base_context_obs_learned_dominance_basic_perdim_trainmax_train_to_test +pairwise_weight: 1.0 +ridge_lambdas: 0,0.01,0.1,1,10,100 +selector_chart_feature_mode: base_context_obs_obj +source_index: None +success_bonus: 1.0 +target: utility_margin +threshold_scope: global diff --git a/workspace/runs/ctt_base_context_obs_learned_dominance_basic_perdim_trainmax_train_to_test/data_hash.txt b/workspace/runs/ctt_base_context_obs_learned_dominance_basic_perdim_trainmax_train_to_test/data_hash.txt new file mode 100644 index 0000000000000000000000000000000000000000..d91fbacd145085f2d9a143f07cf1791ff492f9fc --- /dev/null +++ b/workspace/runs/ctt_base_context_obs_learned_dominance_basic_perdim_trainmax_train_to_test/data_hash.txt @@ -0,0 +1,6 @@ +{ + "calibration_input": "130d86b8c5ba5dc901ee48835454693bf42c361803c01f6b2d80590cb2493491", + "calibration_target_index": "67973c50072aa87a3b97533b6292b08a8b4d88a95571dc22e91b6e8fd19e0d85", + "eval_input": "88d4bfe8c95f8d1579cbe59096cf665dc5f2dfdd99481a386a25e1f3cfa2b09d", + "eval_target_index": "dee1b5225485df6624747dd4c5074bcf514685b54017d05e698d2051638fbf36" +} diff --git a/workspace/runs/ctt_base_context_obs_learned_dominance_basic_perdim_trainmax_train_to_test/eval.log b/workspace/runs/ctt_base_context_obs_learned_dominance_basic_perdim_trainmax_train_to_test/eval.log new file mode 100644 index 0000000000000000000000000000000000000000..58e60527653c049a1edc75cf9002db1286feb6ae --- /dev/null +++ b/workspace/runs/ctt_base_context_obs_learned_dominance_basic_perdim_trainmax_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_perdim_trainmax_rollout_comparison/combined_measured_candidates.json +num_eval_rows=144 diff --git a/workspace/runs/ctt_base_context_obs_learned_dominance_basic_perdim_trainmax_train_to_test/git_hash.txt b/workspace/runs/ctt_base_context_obs_learned_dominance_basic_perdim_trainmax_train_to_test/git_hash.txt new file mode 100644 index 0000000000000000000000000000000000000000..293c6ad4e775f2f937dca574f1b83a5f93f0bdb1 --- /dev/null +++ b/workspace/runs/ctt_base_context_obs_learned_dominance_basic_perdim_trainmax_train_to_test/git_hash.txt @@ -0,0 +1 @@ +14e81292e2004925f81e0bb4810f6f144bfd87b7 diff --git a/workspace/runs/ctt_base_context_obs_learned_dominance_basic_perdim_trainmax_train_to_test/metrics.json b/workspace/runs/ctt_base_context_obs_learned_dominance_basic_perdim_trainmax_train_to_test/metrics.json new file mode 100644 index 0000000000000000000000000000000000000000..bd947d17bffaf4b434996e579852fcc12f456512 --- /dev/null +++ b/workspace/runs/ctt_base_context_obs_learned_dominance_basic_perdim_trainmax_train_to_test/metrics.json @@ -0,0 +1,8879 @@ +{ + "calibration_input": "runs/ctt_base_context_obs_train_cal_perdim_trainmax_rollout_comparison/combined_measured_candidates.json", + "calibration_model_selection": { + "base_success": 0.2986111111111111, + "base_utility": 0.748683326644823, + "coverage": 0.15625, + "fallback_rate": 0.84375, + "hidden_chart_oracle_success": 0.6597222222222222, + "hidden_chart_oracle_utility": 1.4150809271054134, + "outcome_ptr": 0.2638888888888889, + "proposal_oracle_success": 0.1909722222222222, + "proposal_oracle_utility": 0.546237602631057, + "selected_success": 0.3020833333333333, + "selected_success_gain_over_base": 0.003472222222222222, + "selected_utility": 0.7507154924161215, + "selector_regret": 0.07099821336386311, + "success_selector_gap": 0.03819444444444445, + "success_support_gap": 0.4826388888888889, + "support_gap": 0.88937310987482 + }, + "calibration_summary": { + "base_success": 0.2986111111111111, + "base_utility": 0.748683326644823, + "coverage": 0.15625, + "fallback_rate": 0.84375, + "hidden_chart_oracle_success": 0.6597222222222222, + "hidden_chart_oracle_utility": 1.4150809271054134, + "outcome_ptr": 0.2638888888888889, + "pairwise_causal_calibration_accuracy": 0.48271204754186925, + "pairwise_causal_calibration_confidence": 0.5195073990636564, + "pairwise_causal_calibration_ece": 0.036795351521787284, + "pairwise_causal_calibration_pairs": 7404.0, + "proposal_oracle_success": 0.1909722222222222, + "proposal_oracle_utility": 0.546237602631057, + "selected_success": 0.3020833333333333, + "selected_success_gain_over_base": 0.003472222222222222, + "selected_utility": 0.7507154924161215, + "selector_regret": 0.07099821336386311, + "success_selector_gap": 0.03819444444444445, + "success_support_gap": 0.4826388888888889, + "support_gap": 0.88937310987482 + }, + "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_perdim_trainmax_rollout_comparison/combined_measured_candidates.json", + "eval_summary": { + "base_success": 0.3333333333333333, + "base_utility": 0.8338337515791258, + "coverage": 0.2361111111111111, + "fallback_rate": 0.7638888888888888, + "hidden_chart_oracle_success": 0.7291666666666666, + "hidden_chart_oracle_utility": 1.5183161981403828, + "outcome_ptr": 0.3611111111111111, + "pairwise_causal_calibration_accuracy": 0.5289459311851448, + "pairwise_causal_calibration_confidence": 0.5203291819836151, + "pairwise_causal_calibration_ece": 0.010301400843750107, + "pairwise_causal_calibration_pairs": 3662.0, + "proposal_oracle_success": 0.2222222222222222, + "proposal_oracle_utility": 0.6232009150553495, + "selected_success": 0.3055555555555556, + "selected_success_gain_over_base": -0.027777777777777776, + "selected_utility": 0.7862164272616307, + "selector_regret": 0.12719054139840105, + "success_selector_gap": 0.06944444444444445, + "success_support_gap": 0.5138888888888888, + "support_gap": 0.9091424554741631 + }, + "eval_target_content_hash": "0c38d73dc9bc4ab756e12006671b5d12f1a23358b281e54b3f6a9ae3cbf6e9af", + "eval_target_split_hash": "84151df9e88c219b76751677558cc5bbbc19082bf358cea7330158e9e403e25f", + "feature_mean": [ + 0.0, + 4.182546837171281, + 1.8836221616495297, + -1.0601087908747155e-18, + 3.5, + 0.125, + 1.2092013888888888, + 0.7583173435164257, + 1.9059527259604592, + 8.0 + ], + "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" + ], + "feature_set": "basic", + "feature_std": [ + 1.0, + 2.4903288136152946, + 2.1478743840800227, + 0.999998700632758, + 2.29128884747792, + 0.3307199138830738, + 1.490882562848811, + 0.3689689239790769, + 0.7063143432867478, + 1e-06 + ], + "fit_design": { + "fit_objective": "pointwise", + "num_candidate_rows": 2304, + "num_fit_rows": 2304, + "num_pairwise_rows": 0, + "num_pointwise_rows": 2304, + "pairwise_weight": 0.0 + }, + "fit_objective": "pointwise", + "k": 8, + "num_calibration_candidates": 2304, + "num_calibration_rows": 288, + "num_eval_candidates": 1152, + "num_eval_rows": 144, + "pairwise_causal_calibration": { + "calibration": { + "accuracy": 0.48271204754186925, + "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": 0.03614093339883573, + "accuracy": 0.4831415030467163, + "confidence": 0.519282436445552, + "count": 7385, + "lower": 0.5, + "upper": 0.6 + }, + { + "abs_gap": 0.29115734299532703, + "accuracy": 0.3157894736842105, + "confidence": 0.6069468166795375, + "count": 19, + "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": 0.036795351521787284, + "mean_confidence": 0.5195073990636564, + "n_bins": 10, + "num_pairs": 7404, + "num_rows": 288 + }, + "eval": { + "accuracy": 0.5289459311851448, + "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": 0.00948497610430099, + "accuracy": 0.5295728368017525, + "confidence": 0.5200878606974515, + "count": 3652, + "lower": 0.5, + "upper": 0.6 + }, + { + "abs_gap": 0.3084597156905677, + "accuracy": 0.3, + "confidence": 0.6084597156905677, + "count": 10, + "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": 0.010301400843750107, + "mean_confidence": 0.5203291819836151, + "n_bins": 10, + "num_pairs": 3662, + "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": 0.7282466888427734, + "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": 0.0, + "pairwise_causal_calibration_accuracy": 0.6428571428571429, + "pairwise_causal_calibration_confidence": 0.5059729607314479, + "pairwise_causal_calibration_ece": 0.13688418212569498, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.32231182251355883, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.7277753949165344, + "seed": "2012211287/train0", + "selected_candidate_index": 2, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.7282466888427734, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.2722246050834656, + "task_id": "LiftPegUpright-v1", + "tau": -0.08708576405217451, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.20915235579013824, + "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": 0.5, + "pairwise_causal_calibration_confidence": 0.5359843336125516, + "pairwise_causal_calibration_ece": 0.0359843336125516, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.05797252706199239, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.20280277729034424, + "seed": "307068879/train0", + "selected_candidate_index": 6, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.19984664022922516, + "selected_utility_gain_over_base": -0.009305715560913086, + "selector_regret": 0.0029561370611190796, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7971972227096558, + "task_id": "LiftPegUpright-v1", + "tau": -0.08708576405217451, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.725517988204956, + "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": 0.5, + "pairwise_causal_calibration_confidence": 0.5176761329820321, + "pairwise_causal_calibration_ece": 0.017676132982032078, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.16932960943008368, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.7180102467536926, + "seed": "1073161778/train0", + "selected_candidate_index": 3, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.725517988204956, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.2819897532463074, + "task_id": "LiftPegUpright-v1", + "tau": -0.08708576405217451, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.525360107421875, + "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": 0.39285714285714285, + "pairwise_causal_calibration_confidence": 0.5146805911324133, + "pairwise_causal_calibration_ece": 0.12182344827527042, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.2021012748656509, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.6124615669250488, + "seed": "608016290/train0", + "selected_candidate_index": 7, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.525360107421875, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.08710145950317383, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.3875384330749512, + "task_id": "LiftPegUpright-v1", + "tau": -0.08708576405217451, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.2097676396369934, + "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": 0.0, + "pairwise_causal_calibration_accuracy": 0.6428571428571429, + "pairwise_causal_calibration_confidence": 0.5358891522669097, + "pairwise_causal_calibration_ece": 0.10696799059023321, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.06101528493441861, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.203963503241539, + "seed": "441046872/train0", + "selected_candidate_index": 7, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.203963503241539, + "selected_utility_gain_over_base": -0.005804136395454407, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.796036496758461, + "task_id": "LiftPegUpright-v1", + "tau": -0.08708576405217451, + "train_seed": "0" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-a9ac13384fe5afc835f1", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.35714285714285715, + "pairwise_causal_calibration_confidence": 0.5554844314022709, + "pairwise_causal_calibration_ece": 0.1983415742594138, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.48462039184575295, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.5233199596405029, + "seed": "1522272686/train0", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.13452494144439697, + "selected_utility_gain_over_base": -1.865475058555603, + "selector_regret": 0.38879501819610596, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.476680040359497, + "task_id": "PickCube-v1", + "tau": -0.08708576405217451, + "train_seed": "0" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-06babf453b22ba8f2407", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "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": -0.44234982955420926, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1747178030/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": -0.08708576405217451, + "train_seed": "0" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "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": 0.0, + "pairwise_causal_calibration_accuracy": 0.14285714285714285, + "pairwise_causal_calibration_confidence": 0.505522972236195, + "pairwise_causal_calibration_ece": 0.3626658293790522, + "pairwise_causal_calibration_pairs": 7.0, + "predicted_margin": -0.236132745149595, + "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": -0.08708576405217451, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.22852759063243866, + "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": 0.5714285714285714, + "pairwise_causal_calibration_confidence": 0.5096685000943126, + "pairwise_causal_calibration_ece": 0.06176007133425876, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.3069393465412038, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.05616961047053337, + "seed": "1461069713/train0", + "selected_candidate_index": 3, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.22852759063243866, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.10317284241318703, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.032693684101104736, + "chart_id": "ms-34726202b475d29cdf38", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.25342071056365967, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.8571428571428571, + "pairwise_causal_calibration_confidence": 0.5158477675274574, + "pairwise_causal_calibration_ece": 0.34129508961539967, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.4163488178444213, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.044337254017591476, + "seed": "626486813/train0", + "selected_candidate_index": 2, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.032693684101104736, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.01164356991648674, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.2090834565460682, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "train_seed": "0" + }, + { + "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": 0.39285714285714285, + "pairwise_causal_calibration_confidence": 0.5044044230750341, + "pairwise_causal_calibration_ece": 0.11154728021789123, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.48298713602984866, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.14525407552719116, + "seed": "2054899126/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": 1.0, + "support_gap": 1.8547459244728088, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.10593423992395401, + "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": 0.6428571428571429, + "pairwise_causal_calibration_confidence": 0.5399196041851851, + "pairwise_causal_calibration_ece": 0.10293753867195776, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.008075281416847935, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.05979333445429802, + "seed": "672328262/train0", + "selected_candidate_index": 2, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.045009810477495193, + "selected_utility_gain_over_base": -0.060924429446458817, + "selector_regret": 0.014783523976802826, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.940206665545702, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.19699124991893768, + "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": 0.6785714285714286, + "pairwise_causal_calibration_confidence": 0.518137575224214, + "pairwise_causal_calibration_ece": 0.1604338533472146, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.260557776573205, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.2748139500617981, + "seed": "767747424/train0", + "selected_candidate_index": 4, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.19699124991893768, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.07782270014286041, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.25001442432403564, + "chart_id": "ms-6b789356d65b3312fbae", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.22781728208065033, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5714285714285714, + "pairwise_causal_calibration_confidence": 0.5171374463367512, + "pairwise_causal_calibration_ece": 0.054291125091820214, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.22432480842207286, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.14440619945526123, + "seed": "278008524/train0", + "selected_candidate_index": 6, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.25001442432403564, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0834110826253891, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.11255800724029541, + "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": 0.42857142857142855, + "pairwise_causal_calibration_confidence": 0.526177933898391, + "pairwise_causal_calibration_ece": 0.09760650532696241, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.18030681051640662, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.0428844690322876, + "seed": "1845489013/train0", + "selected_candidate_index": 0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.11255800724029541, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9571155309677124, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.05078810453414917, + "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": 0.03571428571428571, + "pairwise_causal_calibration_confidence": 0.5093073844341804, + "pairwise_causal_calibration_ece": 0.47359309871989475, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.3828326459758136, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.07115759700536728, + "seed": "626486813/train0", + "selected_candidate_index": 3, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.05078810453414917, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.02036949247121811, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.006586454808712006, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "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": 0.5, + "pairwise_causal_calibration_confidence": 0.5130431632983611, + "pairwise_causal_calibration_ece": 0.013043163298361082, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.38185146507477247, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.011163711547851562, + "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": 0.2515503764152527, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.07103653997182846, + "chart_id": "ms-8577816da2917a2bb63d", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_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": 0.5199574926341867, + "pairwise_causal_calibration_ece": 0.23004250736581333, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.1685022894131509, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.16448359191417694, + "seed": "585904130/train0", + "selected_candidate_index": 7, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.07103653997182846, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.09344705194234848, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.835516408085823, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.16783273220062256, + "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": 0.7857142857142857, + "pairwise_causal_calibration_confidence": 0.5131339872831697, + "pairwise_causal_calibration_ece": 0.272580298431116, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.3913587564835576, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.023719489574432373, + "seed": "141984615/train0", + "selected_candidate_index": 6, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.16783273220062256, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.2620577812194824, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.06810259819030762, + "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": 0.5357142857142857, + "pairwise_causal_calibration_confidence": 0.5284760575281217, + "pairwise_causal_calibration_ece": 0.007238228186164042, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.04205563220625658, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.25842753052711487, + "seed": "750448296/train0", + "selected_candidate_index": 6, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.21836097538471222, + "selected_utility_gain_over_base": 0.1502583771944046, + "selector_regret": 0.04006655514240265, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7415724694728851, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.11957359313964844, + "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": 0.6071428571428571, + "pairwise_causal_calibration_confidence": 0.5078235229574811, + "pairwise_causal_calibration_ece": 0.09931933418537597, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.4612658703790792, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.10778365284204483, + "seed": "636680137/train0", + "selected_candidate_index": 5, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.11957359313964844, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.06278067082166672, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.10184377431869507, + "chart_id": "ms-a8731059713697ce5e2a", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.19710752367973328, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.2857142857142857, + "pairwise_causal_calibration_confidence": 0.5107241132601466, + "pairwise_causal_calibration_ece": 0.22500982754586085, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.3637567221004802, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.029627859592437744, + "seed": "767747424/train0", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.10184377431869507, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.16747966408729553, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "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": 0.3076923076923077, + "pairwise_causal_calibration_confidence": 0.5079098226596691, + "pairwise_causal_calibration_ece": 0.2002175149673614, + "pairwise_causal_calibration_pairs": 13.0, + "predicted_margin": -0.2742629480034105, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1375492786/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": -0.08708576405217451, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.09504059702157974, + "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": 1.0, + "pairwise_causal_calibration_accuracy": 0.7857142857142857, + "pairwise_causal_calibration_confidence": 0.5405964043309816, + "pairwise_causal_calibration_ece": 0.24511788138330415, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.064572249902453, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.15247176587581635, + "seed": "2038594191/train0", + "selected_candidate_index": 7, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.15247176587581635, + "selected_utility_gain_over_base": 0.0574311688542366, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8475282341241837, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.21340738236904144, + "chart_id": "ms-b7d7021b782f73816083", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.22398948669433594, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.6071428571428571, + "pairwise_causal_calibration_confidence": 0.5103188681066618, + "pairwise_causal_calibration_ece": 0.09682398903619527, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.4197005205791743, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.13244229555130005, + "seed": "750448296/train0", + "selected_candidate_index": 5, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.21340738236904144, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.09154719114303589, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "train_seed": "0" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-b8c3ec3ea915c5d35b3e", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "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": -0.3834696689792132, + "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": -0.08708576405217451, + "train_seed": "0" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-b8fc1568ce66c45c8d8b", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2786632478237152, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.42857142857142855, + "pairwise_causal_calibration_confidence": 0.5138872754655867, + "pairwise_causal_calibration_ece": 0.08531584689415811, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.3043353554007297, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.19230616092681885, + "seed": "747260437/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": 0.08635708689689636, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "train_seed": "0" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "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": 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": -0.30747629084713, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "278008524/train0", + "selected_candidate_index": 0, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.17249532043933868, + "chart_id": "ms-f5dd1c43ebc4c0cfc3f1", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2974560856819153, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.8571428571428571, + "pairwise_causal_calibration_confidence": 0.519176648228606, + "pairwise_causal_calibration_ece": 0.3379662089142511, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.27807849592968054, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.21791090071201324, + "seed": "1549324805/train0", + "selected_candidate_index": 0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.17249532043933868, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.04541558027267456, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.07954518496990204, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.09442546218633652, + "chart_id": "ms-f9740a08a98cb251b761", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.6785714285714286, + "pairwise_causal_calibration_confidence": 0.5293359255305599, + "pairwise_causal_calibration_ece": 0.14923550304086874, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.11426068280426432, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.03802649304270744, + "seed": "1822210471/train0", + "selected_candidate_index": 5, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.09442546218633652, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9619735069572926, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "train_seed": "0" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "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": 0.0, + "pairwise_causal_calibration_accuracy": 0.14285714285714285, + "pairwise_causal_calibration_confidence": 0.5207682646931396, + "pairwise_causal_calibration_ece": 0.3779111218359968, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.0957895722528751, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.23744209110736847, + "seed": "750448296/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": 1.0, + "support_gap": 1.7625579088926315, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "train_seed": "0" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-05f34a90fa7355262b8e", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.8214285714285714, + "pairwise_causal_calibration_confidence": 0.5045350428116284, + "pairwise_causal_calibration_ece": 0.316893528616943, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.41396662095926473, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.2057151049375534, + "seed": "440577711/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": 1.0, + "support_gap": 1.7942848950624466, + "task_id": "PushCube-v1", + "tau": -0.08708576405217451, + "train_seed": "0" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "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": 0.0, + "pairwise_causal_calibration_accuracy": 0.39285714285714285, + "pairwise_causal_calibration_confidence": 0.5148670429643715, + "pairwise_causal_calibration_ece": 0.12200990010722862, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.5184929051461216, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.21000581979751587, + "seed": "1324642411/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": 1.0, + "support_gap": 1.7899941802024841, + "task_id": "PushCube-v1", + "tau": -0.08708576405217451, + "train_seed": "0" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-54346a8f79bd9dca1dfe", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.8928571428571429, + "pairwise_causal_calibration_confidence": 0.5043835056714361, + "pairwise_causal_calibration_ece": 0.3884736371857068, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.39804398287220066, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1790188787/train0", + "selected_candidate_index": 0, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": -0.08708576405217451, + "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": 0.5, + "pairwise_causal_calibration_confidence": 0.5197191753325973, + "pairwise_causal_calibration_ece": 0.01971917533259726, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.30952529615270524, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.09678153693675995, + "seed": "1261930456/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": 1.0, + "support_gap": 1.90321846306324, + "task_id": "PushCube-v1", + "tau": -0.08708576405217451, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.04073460400104523, + "chart_id": "ms-c2ae2b78ab7891e74c80", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6, + "pairwise_causal_calibration_confidence": 0.5098428001104167, + "pairwise_causal_calibration_ece": 0.09015719988958326, + "pairwise_causal_calibration_pairs": 25.0, + "predicted_margin": -0.426096250608377, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "509430392/train0", + "selected_candidate_index": 2, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.04073460400104523, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.9592653959989548, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": -0.08708576405217451, + "train_seed": "0" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "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": 0.0, + "pairwise_causal_calibration_accuracy": 0.7142857142857143, + "pairwise_causal_calibration_confidence": 0.5090541462607746, + "pairwise_causal_calibration_ece": 0.2052315680249397, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.3484014667432773, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.14178013801574707, + "seed": "1480856810/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": 0.06770022213459015, + "task_id": "PushCube-v1", + "tau": -0.08708576405217451, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.08620692789554596, + "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": 0.0, + "pairwise_causal_calibration_accuracy": 0.7142857142857143, + "pairwise_causal_calibration_confidence": 0.5348410330504886, + "pairwise_causal_calibration_ece": 0.17944468123522572, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.07878378828125482, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.07642464339733124, + "seed": "2061635888/train0", + "selected_candidate_index": 0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.07517768442630768, + "selected_utility_gain_over_base": -0.011029243469238281, + "selector_regret": 0.0012469589710235596, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9235753566026688, + "task_id": "StackCube-v1", + "tau": -0.08708576405217451, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.1440753936767578, + "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": 0.7142857142857143, + "pairwise_causal_calibration_confidence": 0.5355109972920885, + "pairwise_causal_calibration_ece": 0.1787747169936258, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.06928250847505647, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.15739890933036804, + "seed": "655884605/train0", + "selected_candidate_index": 7, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.13916455209255219, + "selected_utility_gain_over_base": -0.0049108415842056274, + "selector_regret": 0.018234357237815857, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.842601090669632, + "task_id": "StackCube-v1", + "tau": -0.08708576405217451, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.11109928786754608, + "chart_id": "ms-36bc1364d2439099e9ed", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.6071428571428571, + "pairwise_causal_calibration_confidence": 0.5435152760722883, + "pairwise_causal_calibration_ece": 0.06362758107056876, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.10543872296223779, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.0740400105714798, + "seed": "1113371261/train0", + "selected_candidate_index": 0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.0680118203163147, + "selected_utility_gain_over_base": -0.043087467551231384, + "selector_regret": 0.0060281902551651, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9259599894285202, + "task_id": "StackCube-v1", + "tau": -0.08708576405217451, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.21426597237586975, + "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": 0.0, + "pairwise_causal_calibration_accuracy": 0.2857142857142857, + "pairwise_causal_calibration_confidence": 0.5228634819661582, + "pairwise_causal_calibration_ece": 0.23714919625187247, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.055414585551092216, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.21425709128379822, + "seed": "1578990937/train0", + "selected_candidate_index": 4, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.21093030273914337, + "selected_utility_gain_over_base": -0.0033356696367263794, + "selector_regret": 0.003326788544654846, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7857429087162018, + "task_id": "StackCube-v1", + "tau": -0.08708576405217451, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.8320798277854919, + "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": 0.4, + "pairwise_causal_calibration_confidence": 0.5230050972773443, + "pairwise_causal_calibration_ece": 0.12300509727734432, + "pairwise_causal_calibration_pairs": 25.0, + "predicted_margin": -0.09419003293777646, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1509980664/train0", + "selected_candidate_index": 0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.8320798277854919, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.167920172214508, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": -0.08708576405217451, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.18430444598197937, + "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": 0.5137063430075536, + "pairwise_causal_calibration_ece": 0.013706343007553579, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.127498319380692, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.5822281837463379, + "seed": "211245842/train0", + "selected_candidate_index": 2, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.18430444598197937, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.3979237377643585, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.417771816253662, + "task_id": "StackCube-v1", + "tau": -0.08708576405217451, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.2284735143184662, + "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": 0.32142857142857145, + "pairwise_causal_calibration_confidence": 0.5208530758222207, + "pairwise_causal_calibration_ece": 0.19942450439364928, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.09583695341622606, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.5684944987297058, + "seed": "1151410346/train0", + "selected_candidate_index": 2, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.2284735143184662, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.3400209844112396, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.4315055012702942, + "task_id": "StackCube-v1", + "tau": -0.08708576405217451, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.084227055311203, + "chart_id": "ms-9dd1ff4c297a81b13071", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5357142857142857, + "pairwise_causal_calibration_confidence": 0.5433225220880885, + "pairwise_causal_calibration_ece": 0.0076082363738028125, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.09214616076360858, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.0628519058227539, + "seed": "1324133122/train0", + "selected_candidate_index": 4, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.05331718921661377, + "selected_utility_gain_over_base": -0.030909866094589233, + "selector_regret": 0.009534716606140137, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.937148094177246, + "task_id": "StackCube-v1", + "tau": -0.08708576405217451, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.7912740707397461, + "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": 0.6071428571428571, + "pairwise_causal_calibration_confidence": 0.5070868155940906, + "pairwise_causal_calibration_ece": 0.10005604154876646, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.23997313317310848, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.7879213094711304, + "seed": "701667486/train0", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.7912740707397461, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.2120786905288696, + "task_id": "StackCube-v1", + "tau": -0.08708576405217451, + "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": 0.35714285714285715, + "pairwise_causal_calibration_confidence": 0.5098254980752558, + "pairwise_causal_calibration_ece": 0.15268264093239864, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.14882605428987036, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "412102563/train0", + "selected_candidate_index": 0, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": -0.08708576405217451, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.8279688954353333, + "chart_id": "ms-e887dbfa0133af46829d", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.6785714285714286, + "pairwise_causal_calibration_confidence": 0.5240420866010768, + "pairwise_causal_calibration_ece": 0.15452934197035184, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.010543651527821427, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.5970457196235657, + "seed": "1999022574/train0", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.5970457196235657, + "selected_utility_gain_over_base": -0.23092317581176758, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.4029542803764343, + "task_id": "StackCube-v1", + "tau": -0.08708576405217451, + "train_seed": "0" + }, + { + "base_success": 0.0, + "base_utility": 0.7282466888427734, + "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": 0.7037037037037037, + "pairwise_causal_calibration_confidence": 0.5090337381829941, + "pairwise_causal_calibration_ece": 0.1946699655207096, + "pairwise_causal_calibration_pairs": 27.0, + "predicted_margin": -0.3136591221973332, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "2012211287/train1", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.7282466888427734, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.2717533111572266, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "LiftPegUpright-v1", + "tau": -0.08708576405217451, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.20915235579013824, + "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": 0.6071428571428571, + "pairwise_causal_calibration_confidence": 0.5411752591475067, + "pairwise_causal_calibration_ece": 0.06596759799535039, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.018932161107400383, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.20222051441669464, + "seed": "307068879/train1", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.20222051441669464, + "selected_utility_gain_over_base": -0.0069318413734436035, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7977794855833054, + "task_id": "LiftPegUpright-v1", + "tau": -0.08708576405217451, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.725517988204956, + "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": 0.5357142857142857, + "pairwise_causal_calibration_confidence": 0.5189946754984194, + "pairwise_causal_calibration_ece": 0.01671961021586632, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.17924265036524914, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.7134289145469666, + "seed": "1073161778/train1", + "selected_candidate_index": 6, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.725517988204956, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.2865710854530334, + "task_id": "LiftPegUpright-v1", + "tau": -0.08708576405217451, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.525360107421875, + "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": 0.4642857142857143, + "pairwise_causal_calibration_confidence": 0.5137128125233418, + "pairwise_causal_calibration_ece": 0.049427098237627476, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.22391340049944683, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.6101007461547852, + "seed": "608016290/train1", + "selected_candidate_index": 3, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.525360107421875, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.08474063873291016, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.3898992538452148, + "task_id": "LiftPegUpright-v1", + "tau": -0.08708576405217451, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.2097676396369934, + "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": 0.0, + "pairwise_causal_calibration_accuracy": 0.75, + "pairwise_causal_calibration_confidence": 0.5366361005155011, + "pairwise_causal_calibration_ece": 0.21336389948449885, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.018316727382166484, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.20609354972839355, + "seed": "441046872/train1", + "selected_candidate_index": 7, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.20609354972839355, + "selected_utility_gain_over_base": -0.0036740899085998535, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7939064502716064, + "task_id": "LiftPegUpright-v1", + "tau": -0.08708576405217451, + "train_seed": "1" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-a9ac13384fe5afc835f1", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.07142857142857142, + "pairwise_causal_calibration_confidence": 0.5480190882988387, + "pairwise_causal_calibration_ece": 0.4765905168702673, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.23140829517862893, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.5422300696372986, + "seed": "1522272686/train1", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.09935368597507477, + "selected_utility_gain_over_base": -1.9006463140249252, + "selector_regret": 0.4428763836622238, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.4577699303627014, + "task_id": "PickCube-v1", + "tau": -0.08708576405217451, + "train_seed": "1" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-06babf453b22ba8f2407", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "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": -0.5995592153240683, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1747178030/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": -0.08708576405217451, + "train_seed": "1" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "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": 0.0, + "pairwise_causal_calibration_accuracy": 0.25, + "pairwise_causal_calibration_confidence": 0.5141318477854203, + "pairwise_causal_calibration_ece": 0.2641318477854203, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.12042987778329652, + "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": -0.08708576405217451, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.22852759063243866, + "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": 0.5357142857142857, + "pairwise_causal_calibration_confidence": 0.5043374545649476, + "pairwise_causal_calibration_ece": 0.03137683114933809, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.6098197544005363, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.05644957348704338, + "seed": "1461069713/train1", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.22852759063243866, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.10289287939667702, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.032693684101104736, + "chart_id": "ms-34726202b475d29cdf38", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.25342071056365967, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.8571428571428571, + "pairwise_causal_calibration_confidence": 0.5181952953089655, + "pairwise_causal_calibration_ece": 0.33894756183389163, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.45903664260776045, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.06662949174642563, + "seed": "626486813/train1", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.032693684101104736, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.03393580764532089, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.18679121881723404, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "train_seed": "1" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-4cc40587ec490257b1d2", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.25, + "pairwise_causal_calibration_confidence": 0.509379035536545, + "pairwise_causal_calibration_ece": 0.25937903553654496, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.48848503912158514, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.1511060744524002, + "seed": "2054899126/train1", + "selected_candidate_index": 1, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8488939255475998, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.10593423992395401, + "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": 0.35714285714285715, + "pairwise_causal_calibration_confidence": 0.5306629314646217, + "pairwise_causal_calibration_ece": 0.17352007432176458, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.17563335554996382, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.053388457745313644, + "seed": "672328262/train1", + "selected_candidate_index": 6, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.10593423992395401, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9466115422546864, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.19699124991893768, + "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": 0.8571428571428571, + "pairwise_causal_calibration_confidence": 0.5117962012101025, + "pairwise_causal_calibration_ece": 0.3453466559327546, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.5162388991353228, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.2738197147846222, + "seed": "767747424/train1", + "selected_candidate_index": 7, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.19699124991893768, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.07682846486568451, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.25001442432403564, + "chart_id": "ms-6b789356d65b3312fbae", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.22781728208065033, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.4642857142857143, + "pairwise_causal_calibration_confidence": 0.517634766627836, + "pairwise_causal_calibration_ece": 0.05334905234212173, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.27049207485774296, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.24920666217803955, + "seed": "278008524/train1", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.25001442432403564, + "selected_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": -0.08708576405217451, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.11255800724029541, + "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": 0.6785714285714286, + "pairwise_causal_calibration_confidence": 0.5244266428569392, + "pairwise_causal_calibration_ece": 0.15414478571448942, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.1558320961817238, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.09529292583465576, + "seed": "1845489013/train1", + "selected_candidate_index": 5, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.11255800724029541, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9047070741653442, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.05078810453414917, + "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": 0.32142857142857145, + "pairwise_causal_calibration_confidence": 0.5097899988852159, + "pairwise_causal_calibration_ece": 0.18836142745664447, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.34827465424844745, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.05724402144551277, + "seed": "626486813/train1", + "selected_candidate_index": 7, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.05078810453414917, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.006455916911363602, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.020500030368566513, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "train_seed": "1" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-7e295b5859c07c90487f", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.26271408796310425, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.42857142857142855, + "pairwise_causal_calibration_confidence": 0.5115540705877472, + "pairwise_causal_calibration_ece": 0.08298264201631861, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.342056434081721, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.013422667980194092, + "seed": "767747424/train1", + "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": 0.24929141998291016, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.07103653997182846, + "chart_id": "ms-8577816da2917a2bb63d", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_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": 0.5194057469796561, + "pairwise_causal_calibration_ece": 0.019405746979656135, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.1881172524610201, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.15698926150798798, + "seed": "585904130/train1", + "selected_candidate_index": 3, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.07103653997182846, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.08595272153615952, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.843010738492012, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.16783273220062256, + "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": 0.2857142857142857, + "pairwise_causal_calibration_confidence": 0.5141218630904353, + "pairwise_causal_calibration_ece": 0.22840757737614958, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.3317204404518946, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.04397805407643318, + "seed": "141984615/train1", + "selected_candidate_index": 5, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.16783273220062256, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.2417992167174816, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.06810259819030762, + "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": 0.32142857142857145, + "pairwise_causal_calibration_confidence": 0.5294099946434728, + "pairwise_causal_calibration_ece": 0.20798142321490137, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.00072938120328149, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.2347063571214676, + "seed": "750448296/train1", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.17607416212558746, + "selected_utility_gain_over_base": 0.10797156393527985, + "selector_regret": 0.05863219499588013, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7652936428785324, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.11957359313964844, + "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": 0.2857142857142857, + "pairwise_causal_calibration_confidence": 0.5062252429740812, + "pairwise_causal_calibration_ece": 0.22051095725979553, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.4800313079785443, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.11253269761800766, + "seed": "636680137/train1", + "selected_candidate_index": 3, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.11957359313964844, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.05803162604570389, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.10184377431869507, + "chart_id": "ms-a8731059713697ce5e2a", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.19710752367973328, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.2857142857142857, + "pairwise_causal_calibration_confidence": 0.5147820985446077, + "pairwise_causal_calibration_ece": 0.22906781283032196, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.4562226540231151, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.04315249249339104, + "seed": "767747424/train1", + "selected_candidate_index": 0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.10184377431869507, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.15395503118634224, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "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": 0.39285714285714285, + "pairwise_causal_calibration_confidence": 0.5110081951001124, + "pairwise_causal_calibration_ece": 0.11815105224296957, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.19548314582621923, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.14743654429912567, + "seed": "1375492786/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": 1.0, + "support_gap": 1.8525634557008743, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.09504059702157974, + "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": 1.0, + "pairwise_causal_calibration_accuracy": 0.32142857142857145, + "pairwise_causal_calibration_confidence": 0.5321023921216932, + "pairwise_causal_calibration_ece": 0.21067382069312174, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.0309942873094304, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.1902807205915451, + "seed": "2038594191/train1", + "selected_candidate_index": 4, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.14092111587524414, + "selected_utility_gain_over_base": 0.0458805188536644, + "selector_regret": 0.049359604716300964, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.809719279408455, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.21340738236904144, + "chart_id": "ms-b7d7021b782f73816083", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.22398948669433594, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.4642857142857143, + "pairwise_causal_calibration_confidence": 0.5110397432238398, + "pairwise_causal_calibration_ece": 0.046754028938125525, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.18528838525882577, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.23850953578948975, + "seed": "750448296/train1", + "selected_candidate_index": 5, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.21340738236904144, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.025102153420448303, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "train_seed": "1" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-b8c3ec3ea915c5d35b3e", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "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": -0.38546952995718986, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "2054899126/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": "PullCube-v1", + "tau": -0.08708576405217451, + "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": 0.8928571428571429, + "pairwise_causal_calibration_confidence": 0.5196244343388416, + "pairwise_causal_calibration_ece": 0.37323270851830126, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.41501670336545593, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.1342230886220932, + "seed": "747260437/train1", + "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": 0.144440159201622, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "train_seed": "1" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "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": 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": -0.2704976647284718, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "278008524/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": -0.08708576405217451, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.17249532043933868, + "chart_id": "ms-f5dd1c43ebc4c0cfc3f1", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2974560856819153, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.7857142857142857, + "pairwise_causal_calibration_confidence": 0.5195775633028077, + "pairwise_causal_calibration_ece": 0.266136722411478, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.22126079522588438, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.3088321089744568, + "seed": "1549324805/train1", + "selected_candidate_index": 0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.17249532043933868, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.1363367885351181, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.09442546218633652, + "chart_id": "ms-f9740a08a98cb251b761", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5714285714285714, + "pairwise_causal_calibration_confidence": 0.5263410515594875, + "pairwise_causal_calibration_ece": 0.04508751986908388, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.1660994968065328, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.07539409399032593, + "seed": "1822210471/train1", + "selected_candidate_index": 5, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.09442546218633652, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.924605906009674, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "train_seed": "1" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "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": 0.0, + "pairwise_causal_calibration_accuracy": 0.39285714285714285, + "pairwise_causal_calibration_confidence": 0.5184186974600289, + "pairwise_causal_calibration_ece": 0.12556155460288604, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.1722838597935084, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.2027529925107956, + "seed": "750448296/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": 1.0, + "support_gap": 1.7972470074892044, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "train_seed": "1" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-05f34a90fa7355262b8e", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.38461538461538464, + "pairwise_causal_calibration_confidence": 0.5045702963969247, + "pairwise_causal_calibration_ece": 0.11995491178154011, + "pairwise_causal_calibration_pairs": 13.0, + "predicted_margin": -0.549285469979439, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "440577711/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": -0.08708576405217451, + "train_seed": "1" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "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": 0.0, + "pairwise_causal_calibration_accuracy": 0.48148148148148145, + "pairwise_causal_calibration_confidence": 0.5161761768637876, + "pairwise_causal_calibration_ece": 0.03469469538230613, + "pairwise_causal_calibration_pairs": 27.0, + "predicted_margin": -0.38742954134254165, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1324642411/train1", + "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": "PushCube-v1", + "tau": -0.08708576405217451, + "train_seed": "1" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-54346a8f79bd9dca1dfe", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5357142857142857, + "pairwise_causal_calibration_confidence": 0.5065734080456246, + "pairwise_causal_calibration_ece": 0.029140877668661136, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.41064536165598464, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.1862330436706543, + "seed": "1790188787/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": 1.0, + "support_gap": 1.8137669563293457, + "task_id": "PushCube-v1", + "tau": -0.08708576405217451, + "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": 0.32142857142857145, + "pairwise_causal_calibration_confidence": 0.5188298291687808, + "pairwise_causal_calibration_ece": 0.1974012577402094, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.34495061079317235, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.09997129440307617, + "seed": "1261930456/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": 1.0, + "support_gap": 1.9000287055969238, + "task_id": "PushCube-v1", + "tau": -0.08708576405217451, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.04073460400104523, + "chart_id": "ms-c2ae2b78ab7891e74c80", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.7272727272727273, + "pairwise_causal_calibration_confidence": 0.5095947800098043, + "pairwise_causal_calibration_ece": 0.217677947262923, + "pairwise_causal_calibration_pairs": 22.0, + "predicted_margin": -0.42675791108199884, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "509430392/train1", + "selected_candidate_index": 4, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.04073460400104523, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.9592653959989548, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": -0.08708576405217451, + "train_seed": "1" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "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": 0.0, + "pairwise_causal_calibration_accuracy": 0.32142857142857145, + "pairwise_causal_calibration_confidence": 0.5099948857012214, + "pairwise_causal_calibration_ece": 0.18856631427264997, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.6088141163584672, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1480856810/train1", + "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": "PushCube-v1", + "tau": -0.08708576405217451, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.08620692789554596, + "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": 0.75, + "pairwise_causal_calibration_confidence": 0.5355655008647758, + "pairwise_causal_calibration_ece": 0.21443449913522417, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.004872853139381891, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.09046469628810883, + "seed": "2061635888/train1", + "selected_candidate_index": 4, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.08908586204051971, + "selected_utility_gain_over_base": 0.002878934144973755, + "selector_regret": 0.0013788342475891113, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9095353037118912, + "task_id": "StackCube-v1", + "tau": -0.08708576405217451, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.1440753936767578, + "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": 0.42857142857142855, + "pairwise_causal_calibration_confidence": 0.5393068679380685, + "pairwise_causal_calibration_ece": 0.11073543936663993, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.07124935757530594, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.15038728713989258, + "seed": "655884605/train1", + "selected_candidate_index": 6, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.12485501170158386, + "selected_utility_gain_over_base": -0.01922038197517395, + "selector_regret": 0.025532275438308716, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8496127128601074, + "task_id": "StackCube-v1", + "tau": -0.08708576405217451, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.11109928786754608, + "chart_id": "ms-36bc1364d2439099e9ed", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.39285714285714285, + "pairwise_causal_calibration_confidence": 0.5447882952488178, + "pairwise_causal_calibration_ece": 0.15193115239167487, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.014134138393897249, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.09622243046760559, + "seed": "1113371261/train1", + "selected_candidate_index": 4, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.08607026934623718, + "selected_utility_gain_over_base": -0.0250290185213089, + "selector_regret": 0.010152161121368408, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9037775695323944, + "task_id": "StackCube-v1", + "tau": -0.08708576405217451, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.21426597237586975, + "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": 0.7142857142857143, + "pairwise_causal_calibration_confidence": 0.5249253193280673, + "pairwise_causal_calibration_ece": 0.189360394957647, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.07562260772605807, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.21763037145137787, + "seed": "1578990937/train1", + "selected_candidate_index": 4, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.21256697177886963, + "selected_utility_gain_over_base": -0.001699000597000122, + "selector_regret": 0.00506339967250824, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7823696285486221, + "task_id": "StackCube-v1", + "tau": -0.08708576405217451, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.8320798277854919, + "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": 0.56, + "pairwise_causal_calibration_confidence": 0.524259944555156, + "pairwise_causal_calibration_ece": 0.03574005544484404, + "pairwise_causal_calibration_pairs": 25.0, + "predicted_margin": -0.10500215845255088, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1509980664/train1", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.8320798277854919, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.167920172214508, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": -0.08708576405217451, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.18430444598197937, + "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": 0.39285714285714285, + "pairwise_causal_calibration_confidence": 0.5137524311865478, + "pairwise_causal_calibration_ece": 0.12089528832940494, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.25192844397522873, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.2016763538122177, + "seed": "211245842/train1", + "selected_candidate_index": 2, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.18430444598197937, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.017371907830238342, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7983236461877823, + "task_id": "StackCube-v1", + "tau": -0.08708576405217451, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.2284735143184662, + "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": 0.21428571428571427, + "pairwise_causal_calibration_confidence": 0.5245737063651857, + "pairwise_causal_calibration_ece": 0.31028799207947144, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.14552686699164114, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.2439272701740265, + "seed": "1151410346/train1", + "selected_candidate_index": 3, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.2284735143184662, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.015453755855560303, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7560727298259735, + "task_id": "StackCube-v1", + "tau": -0.08708576405217451, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.084227055311203, + "chart_id": "ms-9dd1ff4c297a81b13071", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6071428571428571, + "pairwise_causal_calibration_confidence": 0.5410803984713379, + "pairwise_causal_calibration_ece": 0.06606245867151916, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.006220904061644952, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.0843733698129654, + "seed": "1324133122/train1", + "selected_candidate_index": 7, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.06854432821273804, + "selected_utility_gain_over_base": -0.015682727098464966, + "selector_regret": 0.015829041600227356, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9156266301870346, + "task_id": "StackCube-v1", + "tau": -0.08708576405217451, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.7912740707397461, + "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": 1.0, + "pairwise_causal_calibration_accuracy": 0.5357142857142857, + "pairwise_causal_calibration_confidence": 0.5088979213620758, + "pairwise_causal_calibration_ece": 0.02681636435220991, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.2717920123841143, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.7947455048561096, + "seed": "701667486/train1", + "selected_candidate_index": 2, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.7912740707397461, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0034714341163635254, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.2052544951438904, + "task_id": "StackCube-v1", + "tau": -0.08708576405217451, + "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": 0.75, + "pairwise_causal_calibration_confidence": 0.5128844404712761, + "pairwise_causal_calibration_ece": 0.23711555952872387, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.26977222222104225, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "412102563/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": "StackCube-v1", + "tau": -0.08708576405217451, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.8279688954353333, + "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": 1.0, + "pairwise_causal_calibration_accuracy": 0.39285714285714285, + "pairwise_causal_calibration_confidence": 0.5293282193230285, + "pairwise_causal_calibration_ece": 0.1364710764658857, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.09841891022097132, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1999022574/train1", + "selected_candidate_index": 0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.8279688954353333, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.1720311045646667, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": -0.08708576405217451, + "train_seed": "1" + }, + { + "base_success": 0.0, + "base_utility": 0.7282466888427734, + "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": 0.0, + "pairwise_causal_calibration_accuracy": 0.8571428571428571, + "pairwise_causal_calibration_confidence": 0.5098518787108952, + "pairwise_causal_calibration_ece": 0.3472909784319619, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.21985746212543253, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.7270426154136658, + "seed": "2012211287/train2", + "selected_candidate_index": 4, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.7282466888427734, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.2729573845863342, + "task_id": "LiftPegUpright-v1", + "tau": -0.08708576405217451, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.20915235579013824, + "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": 0.6785714285714286, + "pairwise_causal_calibration_confidence": 0.536058056899533, + "pairwise_causal_calibration_ece": 0.1425133716718956, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.0032027848618630603, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.2035302072763443, + "seed": "307068879/train2", + "selected_candidate_index": 4, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.2035302072763443, + "selected_utility_gain_over_base": -0.005622148513793945, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7964697927236557, + "task_id": "LiftPegUpright-v1", + "tau": -0.08708576405217451, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.725517988204956, + "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": 1.0, + "pairwise_causal_calibration_accuracy": 0.4642857142857143, + "pairwise_causal_calibration_confidence": 0.5177030926793956, + "pairwise_causal_calibration_ece": 0.053417378393681325, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.18893577023738622, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1073161778/train2", + "selected_candidate_index": 5, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.725517988204956, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.274482011795044, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "LiftPegUpright-v1", + "tau": -0.08708576405217451, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.525360107421875, + "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": 0.6071428571428571, + "pairwise_causal_calibration_confidence": 0.5159644162835159, + "pairwise_causal_calibration_ece": 0.09117844085934124, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.1967433702484556, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.5705108642578125, + "seed": "608016290/train2", + "selected_candidate_index": 3, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.525360107421875, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0451507568359375, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.4294891357421875, + "task_id": "LiftPegUpright-v1", + "tau": -0.08708576405217451, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.2097676396369934, + "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": 0.0, + "pairwise_causal_calibration_accuracy": 0.6785714285714286, + "pairwise_causal_calibration_confidence": 0.5347103383915133, + "pairwise_causal_calibration_ece": 0.14386109017991533, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.037788380273409077, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.2042645812034607, + "seed": "441046872/train2", + "selected_candidate_index": 6, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.2042645812034607, + "selected_utility_gain_over_base": -0.005503058433532715, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7957354187965393, + "task_id": "LiftPegUpright-v1", + "tau": -0.08708576405217451, + "train_seed": "2" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-a9ac13384fe5afc835f1", + "coverage": 1.0, + "execute_generated": 1.0, + "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": 0.5381827836063698, + "pairwise_causal_calibration_ece": 0.0024684978920841427, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.12607764557193257, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1522272686/train2", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.5520380139350891, + "selected_utility_gain_over_base": -1.4479619860649109, + "selector_regret": 1.4479619860649109, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PickCube-v1", + "tau": -0.08708576405217451, + "train_seed": "2" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-06babf453b22ba8f2407", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "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": -0.5527239930103893, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1747178030/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": "PullCube-v1", + "tau": -0.08708576405217451, + "train_seed": "2" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "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": 0.0, + "pairwise_causal_calibration_accuracy": 0.6428571428571429, + "pairwise_causal_calibration_confidence": 0.5099851077722589, + "pairwise_causal_calibration_ece": 0.132872035084884, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.2663815962426988, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "767747424/train2", + "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": -0.08708576405217451, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.22852759063243866, + "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": 0.5, + "pairwise_causal_calibration_confidence": 0.5107514653860995, + "pairwise_causal_calibration_ece": 0.010751465386099501, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.30833078189061486, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.11542954295873642, + "seed": "1461069713/train2", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.22852759063243866, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.04391290992498398, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.032693684101104736, + "chart_id": "ms-34726202b475d29cdf38", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.25342071056365967, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.35714285714285715, + "pairwise_causal_calibration_confidence": 0.5197313409570276, + "pairwise_causal_calibration_ece": 0.1625884838141704, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.24517518114817152, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.10755517333745956, + "seed": "626486813/train2", + "selected_candidate_index": 3, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.032693684101104736, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.07486148923635483, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.1458655372262001, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "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": 0.6071428571428571, + "pairwise_causal_calibration_confidence": 0.5065285020074936, + "pairwise_causal_calibration_ece": 0.10061435513536354, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.37174144193739117, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.13674892485141754, + "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": 1.0, + "support_gap": 1.8632510751485825, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.10593423992395401, + "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": 0.5, + "pairwise_causal_calibration_confidence": 0.5327362714264482, + "pairwise_causal_calibration_ece": 0.032736271426448194, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.014732418206182048, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.07216622680425644, + "seed": "672328262/train2", + "selected_candidate_index": 6, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.05270739272236824, + "selected_utility_gain_over_base": -0.05322684720158577, + "selector_regret": 0.0194588340818882, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9278337731957436, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.19699124991893768, + "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": 0.21428571428571427, + "pairwise_causal_calibration_confidence": 0.5135017084190505, + "pairwise_causal_calibration_ece": 0.2992159941333362, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.37588041949606843, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.25252875685691833, + "seed": "767747424/train2", + "selected_candidate_index": 0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.19699124991893768, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.05553750693798065, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.25001442432403564, + "chart_id": "ms-6b789356d65b3312fbae", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.22781728208065033, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.32142857142857145, + "pairwise_causal_calibration_confidence": 0.5156802850563914, + "pairwise_causal_calibration_ece": 0.19425171362781996, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.2608437162571345, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.31406447291374207, + "seed": "278008524/train2", + "selected_candidate_index": 2, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.25001442432403564, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.06405004858970642, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.11255800724029541, + "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": 0.75, + "pairwise_causal_calibration_confidence": 0.5261804541244063, + "pairwise_causal_calibration_ece": 0.22381954587559372, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.15437862688701337, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.0861951932311058, + "seed": "1845489013/train2", + "selected_candidate_index": 3, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.11255800724029541, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9138048067688942, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.05078810453414917, + "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": 0.7142857142857143, + "pairwise_causal_calibration_confidence": 0.506293731598927, + "pairwise_causal_calibration_ece": 0.2079919826867873, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.2888739023462078, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.05257529020309448, + "seed": "626486813/train2", + "selected_candidate_index": 7, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.05078810453414917, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0017871856689453125, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.025168761610984802, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "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": 0.75, + "pairwise_causal_calibration_confidence": 0.5130065129443258, + "pairwise_causal_calibration_ece": 0.23699348705567425, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.4299004252420342, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.01747530698776245, + "seed": "767747424/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": 0.2452387809753418, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.07103653997182846, + "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": 0.42857142857142855, + "pairwise_causal_calibration_confidence": 0.5199065927971106, + "pairwise_causal_calibration_ece": 0.09133516422568205, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.014285494888099215, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.2938849627971649, + "seed": "585904130/train2", + "selected_candidate_index": 5, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.19301573932170868, + "selected_utility_gain_over_base": 0.12197919934988022, + "selector_regret": 0.10086922347545624, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.706115037202835, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.16783273220062256, + "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": 0.4642857142857143, + "pairwise_causal_calibration_confidence": 0.5067099785179769, + "pairwise_causal_calibration_ece": 0.04242426423226264, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.1614581645625799, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.04447716474533081, + "seed": "141984615/train2", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.16783273220062256, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.24130010604858398, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.06810259819030762, + "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": 0.2857142857142857, + "pairwise_causal_calibration_confidence": 0.5276651423695816, + "pairwise_causal_calibration_ece": 0.24195085665529592, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.05985619371239578, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.4033660888671875, + "seed": "750448296/train2", + "selected_candidate_index": 7, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.23036037385463715, + "selected_utility_gain_over_base": 0.16225777566432953, + "selector_regret": 0.17300571501255035, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.5966339111328125, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.11957359313964844, + "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": 1.0, + "pairwise_causal_calibration_accuracy": 0.75, + "pairwise_causal_calibration_confidence": 0.5088107099176903, + "pairwise_causal_calibration_ece": 0.24118929008230972, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.42177891564909986, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.1247115358710289, + "seed": "636680137/train2", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.11957359313964844, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.005137942731380463, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.04585278779268265, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.10184377431869507, + "chart_id": "ms-a8731059713697ce5e2a", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.19710752367973328, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.35714285714285715, + "pairwise_causal_calibration_confidence": 0.5108345642411548, + "pairwise_causal_calibration_ece": 0.15369170709829766, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.3552047492178657, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.053976815193891525, + "seed": "767747424/train2", + "selected_candidate_index": 7, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.10184377431869507, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.14313070848584175, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "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": 0.6428571428571429, + "pairwise_causal_calibration_confidence": 0.5145465188512656, + "pairwise_causal_calibration_ece": 0.12831062400587734, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.2954350061136046, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.19452734291553497, + "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": 1.0, + "support_gap": 1.805472657084465, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.09504059702157974, + "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": 1.0, + "pairwise_causal_calibration_accuracy": 0.5357142857142857, + "pairwise_causal_calibration_confidence": 0.5393910100975016, + "pairwise_causal_calibration_ece": 0.003676724383215868, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.09610840435854003, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.24880830943584442, + "seed": "2038594191/train2", + "selected_candidate_index": 7, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.23134692013263702, + "selected_utility_gain_over_base": 0.13630632311105728, + "selector_regret": 0.017461389303207397, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7511916905641556, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.21340738236904144, + "chart_id": "ms-b7d7021b782f73816083", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.22398948669433594, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.42857142857142855, + "pairwise_causal_calibration_confidence": 0.5106625601437053, + "pairwise_causal_calibration_ece": 0.08209113157227671, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.32492189510139874, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.17124910652637482, + "seed": "750448296/train2", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.21340738236904144, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.05274038016796112, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "train_seed": "2" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-b8c3ec3ea915c5d35b3e", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "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": -0.46341174706763977, + "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": -0.08708576405217451, + "train_seed": "2" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-b8fc1568ce66c45c8d8b", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2786632478237152, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5357142857142857, + "pairwise_causal_calibration_confidence": 0.5073099404164519, + "pairwise_causal_calibration_ece": 0.028404345297833755, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.41403821961369863, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.18527252972126007, + "seed": "747260437/train2", + "selected_candidate_index": 2, + "selected_success": 1.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 2.0, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.09339071810245514, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "train_seed": "2" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "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": 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": -0.36704685501851353, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "278008524/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": "PullCube-v1", + "tau": -0.08708576405217451, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.17249532043933868, + "chart_id": "ms-f5dd1c43ebc4c0cfc3f1", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2974560856819153, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.7857142857142857, + "pairwise_causal_calibration_confidence": 0.5151424475596346, + "pairwise_causal_calibration_ece": 0.27057183815465113, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.39631033691994494, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.21403594315052032, + "seed": "1549324805/train2", + "selected_candidate_index": 5, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.17249532043933868, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.04154062271118164, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.08342014253139496, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.09442546218633652, + "chart_id": "ms-f9740a08a98cb251b761", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.6428571428571429, + "pairwise_causal_calibration_confidence": 0.5328646888714765, + "pairwise_causal_calibration_ece": 0.10999245398566637, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.12061726254527255, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.058819372206926346, + "seed": "1822210471/train2", + "selected_candidate_index": 3, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.09442546218633652, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9411806277930737, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "train_seed": "2" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "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": 0.0, + "pairwise_causal_calibration_accuracy": 0.39285714285714285, + "pairwise_causal_calibration_confidence": 0.5201858140741172, + "pairwise_causal_calibration_ece": 0.1273286712169744, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.000988532450897639, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.1698276251554489, + "seed": "750448296/train2", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.11492816358804703, + "selected_utility_gain_over_base": -1.885071836411953, + "selector_regret": 0.054899461567401886, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.830172374844551, + "task_id": "PullCube-v1", + "tau": -0.08708576405217451, + "train_seed": "2" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-05f34a90fa7355262b8e", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "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": -0.3642232105295439, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "440577711/train2", + "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": -0.08708576405217451, + "train_seed": "2" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "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": 0.0, + "pairwise_causal_calibration_accuracy": 0.3333333333333333, + "pairwise_causal_calibration_confidence": 0.5164803540872519, + "pairwise_causal_calibration_ece": 0.1831470207539186, + "pairwise_causal_calibration_pairs": 18.0, + "predicted_margin": -0.17737669925866156, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1324642411/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": "PushCube-v1", + "tau": -0.08708576405217451, + "train_seed": "2" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "chart_id": "ms-54346a8f79bd9dca1dfe", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.8636363636363636, + "pairwise_causal_calibration_confidence": 0.5093146507839553, + "pairwise_causal_calibration_ece": 0.3543217128524083, + "pairwise_causal_calibration_pairs": 22.0, + "predicted_margin": -0.37508336216163984, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "1790188787/train2", + "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": -0.08708576405217451, + "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": 0.5714285714285714, + "pairwise_causal_calibration_confidence": 0.5214267898957143, + "pairwise_causal_calibration_ece": 0.05000178153285706, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.3104603102303952, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.08545441925525665, + "seed": "1261930456/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": 1.0, + "support_gap": 1.9145455807447433, + "task_id": "PushCube-v1", + "tau": -0.08708576405217451, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.04073460400104523, + "chart_id": "ms-c2ae2b78ab7891e74c80", + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6818181818181818, + "pairwise_causal_calibration_confidence": 0.5074631947694757, + "pairwise_causal_calibration_ece": 0.1743549870487061, + "pairwise_causal_calibration_pairs": 22.0, + "predicted_margin": -0.32887303074558316, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "509430392/train2", + "selected_candidate_index": 1, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.04073460400104523, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.9592653959989548, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "PushCube-v1", + "tau": -0.08708576405217451, + "train_seed": "2" + }, + { + "base_success": 1.0, + "base_utility": 2.0, + "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": 0.0, + "pairwise_causal_calibration_accuracy": 0.5357142857142857, + "pairwise_causal_calibration_confidence": 0.5037911655250441, + "pairwise_causal_calibration_ece": 0.031923120189241594, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.41400203151287207, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.19622710347175598, + "seed": "1480856810/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": 0.013253256678581238, + "task_id": "PushCube-v1", + "tau": -0.08708576405217451, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.08620692789554596, + "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": 0.0, + "pairwise_causal_calibration_accuracy": 0.42857142857142855, + "pairwise_causal_calibration_confidence": 0.5378360101559639, + "pairwise_causal_calibration_ece": 0.10926458158453539, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.061471356218488246, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.07902413606643677, + "seed": "2061635888/train2", + "selected_candidate_index": 2, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.07194386422634125, + "selected_utility_gain_over_base": -0.014263063669204712, + "selector_regret": 0.00708027184009552, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9209758639335632, + "task_id": "StackCube-v1", + "tau": -0.08708576405217451, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.1440753936767578, + "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": 0.5402960667817187, + "pairwise_causal_calibration_ece": 0.04029606678171871, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.0047442677921717205, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.17242512106895447, + "seed": "655884605/train2", + "selected_candidate_index": 6, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.1574591100215912, + "selected_utility_gain_over_base": 0.013383716344833374, + "selector_regret": 0.014966011047363281, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8275748789310455, + "task_id": "StackCube-v1", + "tau": -0.08708576405217451, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.11109928786754608, + "chart_id": "ms-36bc1364d2439099e9ed", + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.6785714285714286, + "pairwise_causal_calibration_confidence": 0.5452679026603489, + "pairwise_causal_calibration_ece": 0.13330352591107966, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.10321814561223094, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.1003647893667221, + "seed": "1113371261/train2", + "selected_candidate_index": 2, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.09928342700004578, + "selected_utility_gain_over_base": -0.011815860867500305, + "selector_regret": 0.0010813623666763306, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.899635210633278, + "task_id": "StackCube-v1", + "tau": -0.08708576405217451, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.21426597237586975, + "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": 0.5, + "pairwise_causal_calibration_confidence": 0.5243615747594301, + "pairwise_causal_calibration_ece": 0.024361574759430127, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.11122156610391924, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.23259200155735016, + "seed": "1578990937/train2", + "selected_candidate_index": 5, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.21426597237586975, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.018326029181480408, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7674079984426498, + "task_id": "StackCube-v1", + "tau": -0.08708576405217451, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.8320798277854919, + "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": 0.42857142857142855, + "pairwise_causal_calibration_confidence": 0.5212588879913773, + "pairwise_causal_calibration_ece": 0.09268745941994877, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.1890976986089807, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.83889240026474, + "seed": "1509980664/train2", + "selected_candidate_index": 0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.8320798277854919, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.006812572479248047, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.16110759973526, + "task_id": "StackCube-v1", + "tau": -0.08708576405217451, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.18430444598197937, + "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": 0.5176140630711756, + "pairwise_causal_calibration_ece": 0.017614063071175634, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.19743010297696217, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.2142850160598755, + "seed": "211245842/train2", + "selected_candidate_index": 4, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.18430444598197937, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.029980570077896118, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7857149839401245, + "task_id": "StackCube-v1", + "tau": -0.08708576405217451, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.2284735143184662, + "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": 0.39285714285714285, + "pairwise_causal_calibration_confidence": 0.5243490641793181, + "pairwise_causal_calibration_ece": 0.1314919213221753, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.11137550392569598, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.5630103945732117, + "seed": "1151410346/train2", + "selected_candidate_index": 3, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.2284735143184662, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.3345368802547455, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.4369896054267883, + "task_id": "StackCube-v1", + "tau": -0.08708576405217451, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.084227055311203, + "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": 0.6785714285714286, + "pairwise_causal_calibration_confidence": 0.5453211435652117, + "pairwise_causal_calibration_ece": 0.1332502850062169, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.07453570390221465, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.09815146028995514, + "seed": "1324133122/train2", + "selected_candidate_index": 4, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.09103855490684509, + "selected_utility_gain_over_base": 0.00681149959564209, + "selector_regret": 0.007112905383110046, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9018485397100449, + "task_id": "StackCube-v1", + "tau": -0.08708576405217451, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.7912740707397461, + "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": 1.0, + "pairwise_causal_calibration_accuracy": 0.6071428571428571, + "pairwise_causal_calibration_confidence": 0.5112333329328568, + "pairwise_causal_calibration_ece": 0.09590952421000032, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.34033401997115287, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "seed": "701667486/train2", + "selected_candidate_index": 7, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.7912740707397461, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.208725929260254, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "task_id": "StackCube-v1", + "tau": -0.08708576405217451, + "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": 0.25, + "pairwise_causal_calibration_confidence": 0.5173905022150012, + "pairwise_causal_calibration_ece": 0.26739050221500116, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.24386219595665565, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.8275061845779419, + "seed": "412102563/train2", + "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": 1.0, + "support_gap": 1.172493815422058, + "task_id": "StackCube-v1", + "tau": -0.08708576405217451, + "train_seed": "2" + }, + { + "base_success": 0.0, + "base_utility": 0.8279688954353333, + "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": 0.6785714285714286, + "pairwise_causal_calibration_confidence": 0.5292740866826795, + "pairwise_causal_calibration_ece": 0.1492973418887491, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.11507066220959525, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.7970014810562134, + "seed": "1999022574/train2", + "selected_candidate_index": 2, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.8279688954353333, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.2029985189437866, + "task_id": "StackCube-v1", + "tau": -0.08708576405217451, + "train_seed": "2" + } + ], + "schema_version": 1, + "selected_lambda": 0.0, + "selector_calibration_target_content_hash": null, + "selector_calibration_target_split_hash": null, + "selector_chart_feature_mode": null, + "selector_eval_target_content_hash": null, + "selector_eval_target_split_hash": null, + "selector_source_content_hash": null, + "selector_source_split_hash": null, + "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": 0.3958333333333333, + "low": 0.2277777777777778, + "mean": 0.3083333333333334, + "n": 120, + "num_samples": 1000 + }, + "mean": 0.3083333333333334, + "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": 1.0, + "1261930456/train1": 1.0, + "1261930456/train2": 1.0, + "1324133122/train0": 0.0, + "1324133122/train1": 0.0, + "1324133122/train2": 0.0, + "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": 1.0, + "1480856810/train1": 1.0, + "1480856810/train2": 1.0, + "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": 1.0, + "1747178030/train1": 1.0, + "1747178030/train2": 1.0, + "1790188787/train0": 1.0, + "1790188787/train1": 1.0, + "1790188787/train2": 1.0, + "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": 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": 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": 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": 0.0, + "672328262/train1": 0.0, + "672328262/train2": 0.0, + "701667486/train0": 0.0, + "701667486/train1": 0.0, + "701667486/train2": 0.0, + "747260437/train0": 1.0, + "747260437/train1": 1.0, + "747260437/train2": 1.0, + "750448296/train0": 0.3333333333333333, + "750448296/train1": 0.3333333333333333, + "750448296/train2": 0.3333333333333333, + "767747424/train0": 0.5, + "767747424/train1": 0.5, + "767747424/train2": 0.5 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.8053333333333332, + "low": 0.10836363636363637, + "mean": 0.45684848484848484, + "n": 5, + "num_samples": 1000 + }, + "mean": 0.45684848484848484, + "per_task": { + "LiftPegUpright-v1": 0.0, + "PickCube-v1": 1.0, + "PullCube-v1": 0.36, + "PushCube-v1": 0.8333333333333334, + "StackCube-v1": 0.09090909090909091 + } + }, + "metric": "base_success", + "micro": { + "confidence": 0.95, + "high": 0.4097222222222222, + "low": 0.2569444444444444, + "mean": 0.3333333333333333, + "n": 144, + "num_samples": 1000 + }, + "n": 144 + }, + "base_utility": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.9566364926286041, + "low": 0.6609309806488455, + "mean": 0.8028034761858482, + "n": 120, + "num_samples": 1000 + }, + "mean": 0.8028034761858482, + "per_seed": { + "1073161778/train0": 0.725517988204956, + "1073161778/train1": 0.725517988204956, + "1073161778/train2": 0.725517988204956, + "1113371261/train0": 0.11109928786754608, + "1113371261/train1": 0.11109928786754608, + "1113371261/train2": 0.11109928786754608, + "1151410346/train0": 0.2284735143184662, + "1151410346/train1": 0.2284735143184662, + "1151410346/train2": 0.2284735143184662, + "1261930456/train0": 2.0, + "1261930456/train1": 2.0, + "1261930456/train2": 2.0, + "1324133122/train0": 0.084227055311203, + "1324133122/train1": 0.084227055311203, + "1324133122/train2": 0.084227055311203, + "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": 0.16783273220062256, + "141984615/train1": 0.16783273220062256, + "141984615/train2": 0.16783273220062256, + "1461069713/train0": 0.22852759063243866, + "1461069713/train1": 0.22852759063243866, + "1461069713/train2": 0.22852759063243866, + "1480856810/train0": 2.0, + "1480856810/train1": 2.0, + "1480856810/train2": 2.0, + "1509980664/train0": 0.8320798277854919, + "1509980664/train1": 0.8320798277854919, + "1509980664/train2": 0.8320798277854919, + "1522272686/train0": 2.0, + "1522272686/train1": 2.0, + "1522272686/train2": 2.0, + "1549324805/train0": 0.17249532043933868, + "1549324805/train1": 0.17249532043933868, + "1549324805/train2": 0.17249532043933868, + "1578990937/train0": 0.21426597237586975, + "1578990937/train1": 0.21426597237586975, + "1578990937/train2": 0.21426597237586975, + "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": 0.09442546218633652, + "1822210471/train1": 0.09442546218633652, + "1822210471/train2": 0.09442546218633652, + "1845489013/train0": 0.11255800724029541, + "1845489013/train1": 0.11255800724029541, + "1845489013/train2": 0.11255800724029541, + "1999022574/train0": 0.8279688954353333, + "1999022574/train1": 0.8279688954353333, + "1999022574/train2": 0.8279688954353333, + "2012211287/train0": 0.7282466888427734, + "2012211287/train1": 0.7282466888427734, + "2012211287/train2": 0.7282466888427734, + "2038594191/train0": 0.09504059702157974, + "2038594191/train1": 0.09504059702157974, + "2038594191/train2": 0.09504059702157974, + "2054899126/train0": 2.0, + "2054899126/train1": 2.0, + "2054899126/train2": 2.0, + "2061635888/train0": 0.08620692789554596, + "2061635888/train1": 0.08620692789554596, + "2061635888/train2": 0.08620692789554596, + "211245842/train0": 0.18430444598197937, + "211245842/train1": 0.18430444598197937, + "211245842/train2": 0.18430444598197937, + "278008524/train0": 1.1250072121620178, + "278008524/train1": 1.1250072121620178, + "278008524/train2": 1.1250072121620178, + "307068879/train0": 0.20915235579013824, + "307068879/train1": 0.20915235579013824, + "307068879/train2": 0.20915235579013824, + "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": 0.2097676396369934, + "441046872/train1": 0.2097676396369934, + "441046872/train2": 0.2097676396369934, + "509430392/train0": 0.04073460400104523, + "509430392/train1": 0.04073460400104523, + "509430392/train2": 0.04073460400104523, + "585904130/train0": 0.07103653997182846, + "585904130/train1": 0.07103653997182846, + "585904130/train2": 0.07103653997182846, + "608016290/train0": 0.525360107421875, + "608016290/train1": 0.525360107421875, + "608016290/train2": 0.525360107421875, + "626486813/train0": 0.04174089431762695, + "626486813/train1": 0.04174089431762695, + "626486813/train2": 0.04174089431762695, + "636680137/train0": 0.11957359313964844, + "636680137/train1": 0.11957359313964844, + "636680137/train2": 0.11957359313964844, + "655884605/train0": 0.1440753936767578, + "655884605/train1": 0.1440753936767578, + "655884605/train2": 0.1440753936767578, + "672328262/train0": 0.10593423992395401, + "672328262/train1": 0.10593423992395401, + "672328262/train2": 0.10593423992395401, + "701667486/train0": 0.7912740707397461, + "701667486/train1": 0.7912740707397461, + "701667486/train2": 0.7912740707397461, + "747260437/train0": 2.0, + "747260437/train1": 2.0, + "747260437/train2": 2.0, + "750448296/train0": 0.7605033268531164, + "750448296/train1": 0.7605033268531164, + "750448296/train2": 0.7605033268531164, + "767747424/train0": 1.0747087560594082, + "767747424/train1": 1.0747087560594082, + "767747424/train2": 1.0747087560594082 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 1.6347634333101186, + "low": 0.5526382644826715, + "mean": 1.0913353469100866, + "n": 5, + "num_samples": 1000 + }, + "mean": 1.0913353469100866, + "per_task": { + "LiftPegUpright-v1": 0.4796089559793472, + "PickCube-v1": 2.0, + "PullCube-v1": 0.8032506120204925, + "PushCube-v1": 1.6734557673335075, + "StackCube-v1": 0.5003613992170854 + } + }, + "metric": "base_utility", + "micro": { + "confidence": 0.95, + "high": 0.9699174793851044, + "low": 0.6952519603073597, + "mean": 0.8338337515791258, + "n": 144, + "num_samples": 1000 + }, + "n": 144 + }, + "coverage": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.3361111111111111, + "low": 0.18333333333333332, + "mean": 0.2611111111111111, + "n": 120, + "num_samples": 1000 + }, + "mean": 0.2611111111111111, + "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": 0.0, + "1151410346/train1": 0.0, + "1151410346/train2": 0.0, + "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": 0.0, + "1324642411/train1": 0.0, + "1324642411/train2": 0.0, + "1375492786/train0": 0.0, + "1375492786/train1": 0.0, + "1375492786/train2": 0.0, + "141984615/train0": 0.0, + "141984615/train1": 0.0, + "141984615/train2": 0.0, + "1461069713/train0": 0.0, + "1461069713/train1": 0.0, + "1461069713/train2": 0.0, + "1480856810/train0": 0.0, + "1480856810/train1": 0.0, + "1480856810/train2": 0.0, + "1509980664/train0": 0.0, + "1509980664/train1": 0.0, + "1509980664/train2": 0.0, + "1522272686/train0": 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": 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": 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": 0.0, + "2054899126/train1": 0.0, + "2054899126/train2": 0.0, + "2061635888/train0": 1.0, + "2061635888/train1": 1.0, + "2061635888/train2": 1.0, + "211245842/train0": 0.0, + "211245842/train1": 0.0, + "211245842/train2": 0.0, + "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": 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": 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": 1.0, + "655884605/train1": 1.0, + "655884605/train2": 1.0, + "672328262/train0": 1.0, + "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": 0.0, + "767747424/train2": 0.0 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.7175757575757575, + "low": 0.11757575757575758, + "mean": 0.3975757575757576, + "n": 5, + "num_samples": 1000 + }, + "mean": 0.3975757575757576, + "per_task": { + "LiftPegUpright-v1": 0.4, + "PickCube-v1": 1.0, + "PullCube-v1": 0.13333333333333333, + "PushCube-v1": 0.0, + "StackCube-v1": 0.45454545454545453 + } + }, + "metric": "coverage", + "micro": { + "confidence": 0.95, + "high": 0.3055555555555556, + "low": 0.16666666666666666, + "mean": 0.2361111111111111, + "n": 144, + "num_samples": 1000 + }, + "n": 144 + }, + "execute_generated": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.3361111111111111, + "low": 0.18333333333333332, + "mean": 0.2611111111111111, + "n": 120, + "num_samples": 1000 + }, + "mean": 0.2611111111111111, + "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": 0.0, + "1151410346/train1": 0.0, + "1151410346/train2": 0.0, + "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": 0.0, + "1324642411/train1": 0.0, + "1324642411/train2": 0.0, + "1375492786/train0": 0.0, + "1375492786/train1": 0.0, + "1375492786/train2": 0.0, + "141984615/train0": 0.0, + "141984615/train1": 0.0, + "141984615/train2": 0.0, + "1461069713/train0": 0.0, + "1461069713/train1": 0.0, + "1461069713/train2": 0.0, + "1480856810/train0": 0.0, + "1480856810/train1": 0.0, + "1480856810/train2": 0.0, + "1509980664/train0": 0.0, + "1509980664/train1": 0.0, + "1509980664/train2": 0.0, + "1522272686/train0": 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": 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": 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": 0.0, + "2054899126/train1": 0.0, + "2054899126/train2": 0.0, + "2061635888/train0": 1.0, + "2061635888/train1": 1.0, + "2061635888/train2": 1.0, + "211245842/train0": 0.0, + "211245842/train1": 0.0, + "211245842/train2": 0.0, + "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": 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": 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": 1.0, + "655884605/train1": 1.0, + "655884605/train2": 1.0, + "672328262/train0": 1.0, + "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": 0.0, + "767747424/train2": 0.0 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.7175757575757575, + "low": 0.11757575757575758, + "mean": 0.3975757575757576, + "n": 5, + "num_samples": 1000 + }, + "mean": 0.3975757575757576, + "per_task": { + "LiftPegUpright-v1": 0.4, + "PickCube-v1": 1.0, + "PullCube-v1": 0.13333333333333333, + "PushCube-v1": 0.0, + "StackCube-v1": 0.45454545454545453 + } + }, + "metric": "execute_generated", + "micro": { + "confidence": 0.95, + "high": 0.3055555555555556, + "low": 0.16666666666666666, + "mean": 0.2361111111111111, + "n": 144, + "num_samples": 1000 + }, + "n": 144 + }, + "fallback_rate": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.8166666666666667, + "low": 0.663888888888889, + "mean": 0.7388888888888889, + "n": 120, + "num_samples": 1000 + }, + "mean": 0.7388888888888889, + "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": 1.0, + "1151410346/train1": 1.0, + "1151410346/train2": 1.0, + "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": 1.0, + "1324642411/train1": 1.0, + "1324642411/train2": 1.0, + "1375492786/train0": 1.0, + "1375492786/train1": 1.0, + "1375492786/train2": 1.0, + "141984615/train0": 1.0, + "141984615/train1": 1.0, + "141984615/train2": 1.0, + "1461069713/train0": 1.0, + "1461069713/train1": 1.0, + "1461069713/train2": 1.0, + "1480856810/train0": 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": 1.0, + "1549324805/train1": 1.0, + "1549324805/train2": 1.0, + "1578990937/train0": 0.0, + "1578990937/train1": 0.0, + "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": 1.0, + "1999022574/train2": 1.0, + "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": 1.0, + "2054899126/train1": 1.0, + "2054899126/train2": 1.0, + "2061635888/train0": 0.0, + "2061635888/train1": 0.0, + "2061635888/train2": 0.0, + "211245842/train0": 1.0, + "211245842/train1": 1.0, + "211245842/train2": 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": 1.0, + "412102563/train1": 1.0, + "412102563/train2": 1.0, + "440577711/train0": 1.0, + "440577711/train1": 1.0, + "440577711/train2": 1.0, + "441046872/train0": 0.0, + "441046872/train1": 0.0, + "441046872/train2": 0.0, + "509430392/train0": 1.0, + "509430392/train1": 1.0, + "509430392/train2": 1.0, + "585904130/train0": 1.0, + "585904130/train1": 1.0, + "585904130/train2": 0.0, + "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": 0.0, + "655884605/train2": 0.0, + "672328262/train0": 0.0, + "672328262/train1": 1.0, + "672328262/train2": 0.0, + "701667486/train0": 1.0, + "701667486/train1": 1.0, + "701667486/train2": 1.0, + "747260437/train0": 1.0, + "747260437/train1": 1.0, + "747260437/train2": 1.0, + "750448296/train0": 0.6666666666666666, + "750448296/train1": 0.6666666666666666, + "750448296/train2": 0.3333333333333333, + "767747424/train0": 1.0, + "767747424/train1": 1.0, + "767747424/train2": 1.0 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.8824242424242424, + "low": 0.2824242424242424, + "mean": 0.6024242424242424, + "n": 5, + "num_samples": 1000 + }, + "mean": 0.6024242424242424, + "per_task": { + "LiftPegUpright-v1": 0.6, + "PickCube-v1": 0.0, + "PullCube-v1": 0.8666666666666667, + "PushCube-v1": 1.0, + "StackCube-v1": 0.5454545454545454 + } + }, + "metric": "fallback_rate", + "micro": { + "confidence": 0.95, + "high": 0.8333333333333334, + "low": 0.6944444444444444, + "mean": 0.7638888888888888, + "n": 144, + "num_samples": 1000 + }, + "n": 144 + }, + "hidden_chart_oracle_success": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.8555555555555556, + "low": 0.7180555555555556, + "mean": 0.7854166666666668, + "n": 120, + "num_samples": 1000 + }, + "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": 1000 + }, + "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": 1000 + }, + "n": 144 + }, + "hidden_chart_oracle_utility": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 1.7441341616213322, + "low": 1.49932806055165, + "mean": 1.6190625454609595, + "n": 120, + "num_samples": 1000 + }, + "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": 1000 + }, + "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": 1.643800659933024, + "low": 1.3941160800556343, + "mean": 1.5183161981403828, + "n": 144, + "num_samples": 1000 + }, + "n": 144 + }, + "outcome_ptr": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.45138888888888895, + "low": 0.2833333333333333, + "mean": 0.3631944444444445, + "n": 120, + "num_samples": 1000 + }, + "mean": 0.3631944444444445, + "per_seed": { + "1073161778/train0": 0.0, + "1073161778/train1": 0.0, + "1073161778/train2": 1.0, + "1113371261/train0": 0.0, + "1113371261/train1": 0.0, + "1113371261/train2": 0.0, + "1151410346/train0": 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": 0.0, + "1324642411/train1": 0.0, + "1324642411/train2": 0.0, + "1375492786/train0": 0.0, + "1375492786/train1": 0.0, + "1375492786/train2": 0.0, + "141984615/train0": 0.0, + "141984615/train1": 0.0, + "141984615/train2": 0.0, + "1461069713/train0": 0.0, + "1461069713/train1": 0.0, + "1461069713/train2": 0.0, + "1480856810/train0": 0.0, + "1480856810/train1": 0.0, + "1480856810/train2": 0.0, + "1509980664/train0": 1.0, + "1509980664/train1": 1.0, + "1509980664/train2": 1.0, + "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": 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": 1.0, + "1999022574/train2": 0.0, + "2012211287/train0": 0.0, + "2012211287/train1": 1.0, + "2012211287/train2": 0.0, + "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": 1.0, + "2061635888/train2": 0.0, + "211245842/train0": 1.0, + "211245842/train1": 1.0, + "211245842/train2": 1.0, + "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": 0.0, + "440577711/train1": 0.0, + "440577711/train2": 0.0, + "441046872/train0": 0.0, + "441046872/train1": 0.0, + "441046872/train2": 0.0, + "509430392/train0": 1.0, + "509430392/train1": 1.0, + "509430392/train2": 1.0, + "585904130/train0": 1.0, + "585904130/train1": 1.0, + "585904130/train2": 1.0, + "608016290/train0": 1.0, + "608016290/train1": 1.0, + "608016290/train2": 1.0, + "626486813/train0": 1.0, + "626486813/train1": 1.0, + "626486813/train2": 1.0, + "636680137/train0": 0.0, + "636680137/train1": 0.0, + "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": 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": 0.3333333333333333, + "767747424/train0": 0.25, + "767747424/train1": 0.25, + "767747424/train2": 0.25 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.4609696969696969, + "low": 0.1, + "mean": 0.2852121212121212, + "n": 5, + "num_samples": 1000 + }, + "mean": 0.2852121212121212, + "per_task": { + "LiftPegUpright-v1": 0.3333333333333333, + "PickCube-v1": 0.0, + "PullCube-v1": 0.32, + "PushCube-v1": 0.16666666666666666, + "StackCube-v1": 0.6060606060606061 + } + }, + "metric": "outcome_ptr", + "micro": { + "confidence": 0.95, + "high": 0.4444444444444444, + "low": 0.2847222222222222, + "mean": 0.3611111111111111, + "n": 144, + "num_samples": 1000 + }, + "n": 144 + }, + "pairwise_causal_calibration_accuracy": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.5669177270685889, + "low": 0.5090179711731434, + "mean": 0.5385784577810441, + "n": 116, + "num_samples": 1000 + }, + "mean": 0.5385784577810441, + "per_seed": { + "1073161778/train0": 0.5, + "1073161778/train1": 0.5357142857142857, + "1073161778/train2": 0.4642857142857143, + "1113371261/train0": 0.6071428571428571, + "1113371261/train1": 0.39285714285714285, + "1113371261/train2": 0.6785714285714286, + "1151410346/train0": 0.32142857142857145, + "1151410346/train1": 0.21428571428571427, + "1151410346/train2": 0.39285714285714285, + "1261930456/train0": 0.5, + "1261930456/train1": 0.32142857142857145, + "1261930456/train2": 0.5714285714285714, + "1324133122/train0": 0.5357142857142857, + "1324133122/train1": 0.6071428571428571, + "1324133122/train2": 0.6785714285714286, + "1324642411/train0": 0.39285714285714285, + "1324642411/train1": 0.48148148148148145, + "1324642411/train2": 0.3333333333333333, + "1375492786/train0": 0.3076923076923077, + "1375492786/train1": 0.39285714285714285, + "1375492786/train2": 0.6428571428571429, + "141984615/train0": 0.7857142857142857, + "141984615/train1": 0.2857142857142857, + "141984615/train2": 0.4642857142857143, + "1461069713/train0": 0.5714285714285714, + "1461069713/train1": 0.5357142857142857, + "1461069713/train2": 0.5, + "1480856810/train0": 0.7142857142857143, + "1480856810/train1": 0.32142857142857145, + "1480856810/train2": 0.5357142857142857, + "1509980664/train0": 0.4, + "1509980664/train1": 0.56, + "1509980664/train2": 0.42857142857142855, + "1522272686/train0": 0.35714285714285715, + "1522272686/train1": 0.07142857142857142, + "1522272686/train2": 0.5357142857142857, + "1549324805/train0": 0.8571428571428571, + "1549324805/train1": 0.7857142857142857, + "1549324805/train2": 0.7857142857142857, + "1578990937/train0": 0.2857142857142857, + "1578990937/train1": 0.7142857142857143, + "1578990937/train2": 0.5, + "1790188787/train0": 0.8928571428571429, + "1790188787/train1": 0.5357142857142857, + "1790188787/train2": 0.8636363636363636, + "1822210471/train0": 0.6785714285714286, + "1822210471/train1": 0.5714285714285714, + "1822210471/train2": 0.6428571428571429, + "1845489013/train0": 0.42857142857142855, + "1845489013/train1": 0.6785714285714286, + "1845489013/train2": 0.75, + "1999022574/train0": 0.6785714285714286, + "1999022574/train1": 0.39285714285714285, + "1999022574/train2": 0.6785714285714286, + "2012211287/train0": 0.6428571428571429, + "2012211287/train1": 0.7037037037037037, + "2012211287/train2": 0.8571428571428571, + "2038594191/train0": 0.7857142857142857, + "2038594191/train1": 0.32142857142857145, + "2038594191/train2": 0.5357142857142857, + "2054899126/train0": 0.39285714285714285, + "2054899126/train1": 0.25, + "2054899126/train2": 0.6071428571428571, + "2061635888/train0": 0.7142857142857143, + "2061635888/train1": 0.75, + "2061635888/train2": 0.42857142857142855, + "211245842/train0": 0.5, + "211245842/train1": 0.39285714285714285, + "211245842/train2": 0.5, + "278008524/train0": 0.5714285714285714, + "278008524/train1": 0.4642857142857143, + "278008524/train2": 0.32142857142857145, + "307068879/train0": 0.5, + "307068879/train1": 0.6071428571428571, + "307068879/train2": 0.6785714285714286, + "412102563/train0": 0.35714285714285715, + "412102563/train1": 0.75, + "412102563/train2": 0.25, + "440577711/train0": 0.8214285714285714, + "440577711/train1": 0.38461538461538464, + "441046872/train0": 0.6428571428571429, + "441046872/train1": 0.75, + "441046872/train2": 0.6785714285714286, + "509430392/train0": 0.6, + "509430392/train1": 0.7272727272727273, + "509430392/train2": 0.6818181818181818, + "585904130/train0": 0.75, + "585904130/train1": 0.5, + "585904130/train2": 0.42857142857142855, + "608016290/train0": 0.39285714285714285, + "608016290/train1": 0.4642857142857143, + "608016290/train2": 0.6071428571428571, + "626486813/train0": 0.4464285714285714, + "626486813/train1": 0.5892857142857143, + "626486813/train2": 0.5357142857142857, + "636680137/train0": 0.6071428571428571, + "636680137/train1": 0.2857142857142857, + "636680137/train2": 0.75, + "655884605/train0": 0.7142857142857143, + "655884605/train1": 0.42857142857142855, + "655884605/train2": 0.5, + "672328262/train0": 0.6428571428571429, + "672328262/train1": 0.35714285714285715, + "672328262/train2": 0.5, + "701667486/train0": 0.6071428571428571, + "701667486/train1": 0.5357142857142857, + "701667486/train2": 0.6071428571428571, + "747260437/train0": 0.42857142857142855, + "747260437/train1": 0.8928571428571429, + "747260437/train2": 0.5357142857142857, + "750448296/train0": 0.42857142857142855, + "750448296/train1": 0.39285714285714285, + "750448296/train2": 0.369047619047619, + "767747424/train0": 0.4017857142857143, + "767747424/train1": 0.4553571428571429, + "767747424/train2": 0.4910714285714286 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.5720721166995677, + "low": 0.40885414150120036, + "mean": 0.5042714895852151, + "n": 5, + "num_samples": 1000 + }, + "mean": 0.5042714895852151, + "per_task": { + "LiftPegUpright-v1": 0.6016754850088184, + "PickCube-v1": 0.32142857142857145, + "PullCube-v1": 0.5106143856143858, + "PushCube-v1": 0.5693706076059017, + "StackCube-v1": 0.5182683982683982 + } + }, + "metric": "pairwise_causal_calibration_accuracy", + "micro": { + "confidence": 0.95, + "high": 0.5554349934200684, + "low": 0.49485469478006777, + "mean": 0.525911379941231, + "n": 134, + "num_samples": 1000 + }, + "n": 134 + }, + "pairwise_causal_calibration_confidence": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.5231786248652748, + "low": 0.5186922341506015, + "mean": 0.5208649967566025, + "n": 116, + "num_samples": 1000 + }, + "mean": 0.5208649967566025, + "per_seed": { + "1073161778/train0": 0.5176761329820321, + "1073161778/train1": 0.5189946754984194, + "1073161778/train2": 0.5177030926793956, + "1113371261/train0": 0.5435152760722883, + "1113371261/train1": 0.5447882952488178, + "1113371261/train2": 0.5452679026603489, + "1151410346/train0": 0.5208530758222207, + "1151410346/train1": 0.5245737063651857, + "1151410346/train2": 0.5243490641793181, + "1261930456/train0": 0.5197191753325973, + "1261930456/train1": 0.5188298291687808, + "1261930456/train2": 0.5214267898957143, + "1324133122/train0": 0.5433225220880885, + "1324133122/train1": 0.5410803984713379, + "1324133122/train2": 0.5453211435652117, + "1324642411/train0": 0.5148670429643715, + "1324642411/train1": 0.5161761768637876, + "1324642411/train2": 0.5164803540872519, + "1375492786/train0": 0.5079098226596691, + "1375492786/train1": 0.5110081951001124, + "1375492786/train2": 0.5145465188512656, + "141984615/train0": 0.5131339872831697, + "141984615/train1": 0.5141218630904353, + "141984615/train2": 0.5067099785179769, + "1461069713/train0": 0.5096685000943126, + "1461069713/train1": 0.5043374545649476, + "1461069713/train2": 0.5107514653860995, + "1480856810/train0": 0.5090541462607746, + "1480856810/train1": 0.5099948857012214, + "1480856810/train2": 0.5037911655250441, + "1509980664/train0": 0.5230050972773443, + "1509980664/train1": 0.524259944555156, + "1509980664/train2": 0.5212588879913773, + "1522272686/train0": 0.5554844314022709, + "1522272686/train1": 0.5480190882988387, + "1522272686/train2": 0.5381827836063698, + "1549324805/train0": 0.519176648228606, + "1549324805/train1": 0.5195775633028077, + "1549324805/train2": 0.5151424475596346, + "1578990937/train0": 0.5228634819661582, + "1578990937/train1": 0.5249253193280673, + "1578990937/train2": 0.5243615747594301, + "1790188787/train0": 0.5043835056714361, + "1790188787/train1": 0.5065734080456246, + "1790188787/train2": 0.5093146507839553, + "1822210471/train0": 0.5293359255305599, + "1822210471/train1": 0.5263410515594875, + "1822210471/train2": 0.5328646888714765, + "1845489013/train0": 0.526177933898391, + "1845489013/train1": 0.5244266428569392, + "1845489013/train2": 0.5261804541244063, + "1999022574/train0": 0.5240420866010768, + "1999022574/train1": 0.5293282193230285, + "1999022574/train2": 0.5292740866826795, + "2012211287/train0": 0.5059729607314479, + "2012211287/train1": 0.5090337381829941, + "2012211287/train2": 0.5098518787108952, + "2038594191/train0": 0.5405964043309816, + "2038594191/train1": 0.5321023921216932, + "2038594191/train2": 0.5393910100975016, + "2054899126/train0": 0.5044044230750341, + "2054899126/train1": 0.509379035536545, + "2054899126/train2": 0.5065285020074936, + "2061635888/train0": 0.5348410330504886, + "2061635888/train1": 0.5355655008647758, + "2061635888/train2": 0.5378360101559639, + "211245842/train0": 0.5137063430075536, + "211245842/train1": 0.5137524311865478, + "211245842/train2": 0.5176140630711756, + "278008524/train0": 0.5171374463367512, + "278008524/train1": 0.517634766627836, + "278008524/train2": 0.5156802850563914, + "307068879/train0": 0.5359843336125516, + "307068879/train1": 0.5411752591475067, + "307068879/train2": 0.536058056899533, + "412102563/train0": 0.5098254980752558, + "412102563/train1": 0.5128844404712761, + "412102563/train2": 0.5173905022150012, + "440577711/train0": 0.5045350428116284, + "440577711/train1": 0.5045702963969247, + "441046872/train0": 0.5358891522669097, + "441046872/train1": 0.5366361005155011, + "441046872/train2": 0.5347103383915133, + "509430392/train0": 0.5098428001104167, + "509430392/train1": 0.5095947800098043, + "509430392/train2": 0.5074631947694757, + "585904130/train0": 0.5199574926341867, + "585904130/train1": 0.5194057469796561, + "585904130/train2": 0.5199065927971106, + "608016290/train0": 0.5146805911324133, + "608016290/train1": 0.5137128125233418, + "608016290/train2": 0.5159644162835159, + "626486813/train0": 0.5125775759808189, + "626486813/train1": 0.5139926470970907, + "626486813/train2": 0.5130125362779773, + "636680137/train0": 0.5078235229574811, + "636680137/train1": 0.5062252429740812, + "636680137/train2": 0.5088107099176903, + "655884605/train0": 0.5355109972920885, + "655884605/train1": 0.5393068679380685, + "655884605/train2": 0.5402960667817187, + "672328262/train0": 0.5399196041851851, + "672328262/train1": 0.5306629314646217, + "672328262/train2": 0.5327362714264482, + "701667486/train0": 0.5070868155940906, + "701667486/train1": 0.5088979213620758, + "701667486/train2": 0.5112333329328568, + "747260437/train0": 0.5138872754655867, + "747260437/train1": 0.5196244343388416, + "747260437/train2": 0.5073099404164519, + "750448296/train0": 0.5198543967759743, + "750448296/train1": 0.5196228117757805, + "750448296/train2": 0.5195045055291346, + "767747424/train0": 0.5118569560047292, + "767747424/train1": 0.5130660545319694, + "767747424/train2": 0.5118319733441975 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.5371492048512476, + "low": 0.5153984351685331, + "mean": 0.5250404419306169, + "n": 5, + "num_samples": 1000 + }, + "mean": 0.5250404419306169, + "per_task": { + "LiftPegUpright-v1": 0.5229362359705314, + "PickCube-v1": 0.5472287677691599, + "PullCube-v1": 0.5170252389197862, + "PushCube-v1": 0.5109774849646357, + "StackCube-v1": 0.527034482028972 + } + }, + "metric": "pairwise_causal_calibration_confidence", + "micro": { + "confidence": 0.95, + "high": 0.522084890169105, + "low": 0.5180819392343106, + "mean": 0.5200608265890019, + "n": 134, + "num_samples": 1000 + }, + "n": 134 + }, + "pairwise_causal_calibration_ece": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.1556547131521426, + "low": 0.1193992800167444, + "mean": 0.1379737449774899, + "n": 120, + "num_samples": 1000 + }, + "mean": 0.1379737449774899, + "per_seed": { + "1073161778/train0": 0.017676132982032078, + "1073161778/train1": 0.01671961021586632, + "1073161778/train2": 0.053417378393681325, + "1113371261/train0": 0.06362758107056876, + "1113371261/train1": 0.15193115239167487, + "1113371261/train2": 0.13330352591107966, + "1151410346/train0": 0.19942450439364928, + "1151410346/train1": 0.31028799207947144, + "1151410346/train2": 0.1314919213221753, + "1261930456/train0": 0.01971917533259726, + "1261930456/train1": 0.1974012577402094, + "1261930456/train2": 0.05000178153285706, + "1324133122/train0": 0.0076082363738028125, + "1324133122/train1": 0.06606245867151916, + "1324133122/train2": 0.1332502850062169, + "1324642411/train0": 0.12200990010722862, + "1324642411/train1": 0.03469469538230613, + "1324642411/train2": 0.1831470207539186, + "1375492786/train0": 0.2002175149673614, + "1375492786/train1": 0.11815105224296957, + "1375492786/train2": 0.12831062400587734, + "141984615/train0": 0.272580298431116, + "141984615/train1": 0.22840757737614958, + "141984615/train2": 0.04242426423226264, + "1461069713/train0": 0.06176007133425876, + "1461069713/train1": 0.03137683114933809, + "1461069713/train2": 0.010751465386099501, + "1480856810/train0": 0.2052315680249397, + "1480856810/train1": 0.18856631427264997, + "1480856810/train2": 0.031923120189241594, + "1509980664/train0": 0.12300509727734432, + "1509980664/train1": 0.03574005544484404, + "1509980664/train2": 0.09268745941994877, + "1522272686/train0": 0.1983415742594138, + "1522272686/train1": 0.4765905168702673, + "1522272686/train2": 0.0024684978920841427, + "1549324805/train0": 0.3379662089142511, + "1549324805/train1": 0.266136722411478, + "1549324805/train2": 0.27057183815465113, + "1578990937/train0": 0.23714919625187247, + "1578990937/train1": 0.189360394957647, + "1578990937/train2": 0.024361574759430127, + "1747178030/train0": 0.0, + "1747178030/train1": 0.0, + "1747178030/train2": 0.0, + "1790188787/train0": 0.3884736371857068, + "1790188787/train1": 0.029140877668661136, + "1790188787/train2": 0.3543217128524083, + "1822210471/train0": 0.14923550304086874, + "1822210471/train1": 0.04508751986908388, + "1822210471/train2": 0.10999245398566637, + "1845489013/train0": 0.09760650532696241, + "1845489013/train1": 0.15414478571448942, + "1845489013/train2": 0.22381954587559372, + "1999022574/train0": 0.15452934197035184, + "1999022574/train1": 0.1364710764658857, + "1999022574/train2": 0.1492973418887491, + "2012211287/train0": 0.13688418212569498, + "2012211287/train1": 0.1946699655207096, + "2012211287/train2": 0.3472909784319619, + "2038594191/train0": 0.24511788138330415, + "2038594191/train1": 0.21067382069312174, + "2038594191/train2": 0.003676724383215868, + "2054899126/train0": 0.055773640108945616, + "2054899126/train1": 0.12968951776827248, + "2054899126/train2": 0.05030717756768177, + "2061635888/train0": 0.17944468123522572, + "2061635888/train1": 0.21443449913522417, + "2061635888/train2": 0.10926458158453539, + "211245842/train0": 0.013706343007553579, + "211245842/train1": 0.12089528832940494, + "211245842/train2": 0.017614063071175634, + "278008524/train0": 0.027145562545910107, + "278008524/train1": 0.026674526171060864, + "278008524/train2": 0.09712585681390998, + "307068879/train0": 0.0359843336125516, + "307068879/train1": 0.06596759799535039, + "307068879/train2": 0.1425133716718956, + "412102563/train0": 0.15268264093239864, + "412102563/train1": 0.23711555952872387, + "412102563/train2": 0.26739050221500116, + "440577711/train0": 0.316893528616943, + "440577711/train1": 0.11995491178154011, + "440577711/train2": 0.0, + "441046872/train0": 0.10696799059023321, + "441046872/train1": 0.21336389948449885, + "441046872/train2": 0.14386109017991533, + "509430392/train0": 0.09015719988958326, + "509430392/train1": 0.217677947262923, + "509430392/train2": 0.1743549870487061, + "585904130/train0": 0.23004250736581333, + "585904130/train1": 0.019405746979656135, + "585904130/train2": 0.09133516422568205, + "608016290/train0": 0.12182344827527042, + "608016290/train1": 0.049427098237627476, + "608016290/train2": 0.09117844085934124, + "626486813/train0": 0.4074440941676472, + "626486813/train1": 0.263654494645268, + "626486813/train2": 0.18529023325047886, + "636680137/train0": 0.09931933418537597, + "636680137/train1": 0.22051095725979553, + "636680137/train2": 0.24118929008230972, + "655884605/train0": 0.1787747169936258, + "655884605/train1": 0.11073543936663993, + "655884605/train2": 0.04029606678171871, + "672328262/train0": 0.10293753867195776, + "672328262/train1": 0.17352007432176458, + "672328262/train2": 0.032736271426448194, + "701667486/train0": 0.10005604154876646, + "701667486/train1": 0.02681636435220991, + "701667486/train2": 0.09590952421000032, + "747260437/train0": 0.08531584689415811, + "747260437/train1": 0.37323270851830126, + "747260437/train2": 0.028404345297833755, + "750448296/train0": 0.1606577796861187, + "750448296/train1": 0.12676566891863764, + "750448296/train2": 0.15045688648151567, + "767747424/train0": 0.1902881683926222, + "767747424/train1": 0.23038223964120388, + "767747424/train2": 0.205693305843048 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.19086000953021526, + "low": 0.12525600159130484, + "mean": 0.1539728671932507, + "n": 5, + "num_samples": 1000 + }, + "mean": 0.1539728671932507, + "per_task": { + "LiftPegUpright-v1": 0.11584970123844203, + "PickCube-v1": 0.22580019634058843, + "PullCube-v1": 0.149483534146025, + "PushCube-v1": 0.1513149797579122, + "StackCube-v1": 0.12741592448328593 + } + }, + "metric": "pairwise_causal_calibration_ece", + "micro": { + "confidence": 0.95, + "high": 0.16179236050101442, + "low": 0.12588888616028215, + "mean": 0.14274171050098003, + "n": 144, + "num_samples": 1000 + }, + "n": 144 + }, + "pairwise_causal_calibration_pairs": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 26.716666666666665, + "low": 24.608333333333334, + "mean": 25.747916666666665, + "n": 120, + "num_samples": 1000 + }, + "mean": 25.747916666666665, + "per_seed": { + "1073161778/train0": 28.0, + "1073161778/train1": 28.0, + "1073161778/train2": 28.0, + "1113371261/train0": 28.0, + "1113371261/train1": 28.0, + "1113371261/train2": 28.0, + "1151410346/train0": 28.0, + "1151410346/train1": 28.0, + "1151410346/train2": 28.0, + "1261930456/train0": 28.0, + "1261930456/train1": 28.0, + "1261930456/train2": 28.0, + "1324133122/train0": 28.0, + "1324133122/train1": 28.0, + "1324133122/train2": 28.0, + "1324642411/train0": 28.0, + "1324642411/train1": 27.0, + "1324642411/train2": 18.0, + "1375492786/train0": 13.0, + "1375492786/train1": 28.0, + "1375492786/train2": 28.0, + "141984615/train0": 28.0, + "141984615/train1": 28.0, + "141984615/train2": 28.0, + "1461069713/train0": 28.0, + "1461069713/train1": 28.0, + "1461069713/train2": 28.0, + "1480856810/train0": 28.0, + "1480856810/train1": 28.0, + "1480856810/train2": 28.0, + "1509980664/train0": 25.0, + "1509980664/train1": 25.0, + "1509980664/train2": 28.0, + "1522272686/train0": 28.0, + "1522272686/train1": 28.0, + "1522272686/train2": 28.0, + "1549324805/train0": 28.0, + "1549324805/train1": 28.0, + "1549324805/train2": 28.0, + "1578990937/train0": 28.0, + "1578990937/train1": 28.0, + "1578990937/train2": 28.0, + "1747178030/train0": 0.0, + "1747178030/train1": 0.0, + "1747178030/train2": 0.0, + "1790188787/train0": 28.0, + "1790188787/train1": 28.0, + "1790188787/train2": 22.0, + "1822210471/train0": 28.0, + "1822210471/train1": 28.0, + "1822210471/train2": 28.0, + "1845489013/train0": 28.0, + "1845489013/train1": 28.0, + "1845489013/train2": 28.0, + "1999022574/train0": 28.0, + "1999022574/train1": 28.0, + "1999022574/train2": 28.0, + "2012211287/train0": 28.0, + "2012211287/train1": 27.0, + "2012211287/train2": 28.0, + "2038594191/train0": 28.0, + "2038594191/train1": 28.0, + "2038594191/train2": 28.0, + "2054899126/train0": 14.0, + "2054899126/train1": 14.0, + "2054899126/train2": 14.0, + "2061635888/train0": 28.0, + "2061635888/train1": 28.0, + "2061635888/train2": 28.0, + "211245842/train0": 28.0, + "211245842/train1": 28.0, + "211245842/train2": 28.0, + "278008524/train0": 14.0, + "278008524/train1": 14.0, + "278008524/train2": 14.0, + "307068879/train0": 28.0, + "307068879/train1": 28.0, + "307068879/train2": 28.0, + "412102563/train0": 28.0, + "412102563/train1": 28.0, + "412102563/train2": 28.0, + "440577711/train0": 28.0, + "440577711/train1": 13.0, + "440577711/train2": 0.0, + "441046872/train0": 28.0, + "441046872/train1": 28.0, + "441046872/train2": 28.0, + "509430392/train0": 25.0, + "509430392/train1": 22.0, + "509430392/train2": 22.0, + "585904130/train0": 28.0, + "585904130/train1": 28.0, + "585904130/train2": 28.0, + "608016290/train0": 28.0, + "608016290/train1": 28.0, + "608016290/train2": 28.0, + "626486813/train0": 28.0, + "626486813/train1": 28.0, + "626486813/train2": 28.0, + "636680137/train0": 28.0, + "636680137/train1": 28.0, + "636680137/train2": 28.0, + "655884605/train0": 28.0, + "655884605/train1": 28.0, + "655884605/train2": 28.0, + "672328262/train0": 28.0, + "672328262/train1": 28.0, + "672328262/train2": 28.0, + "701667486/train0": 28.0, + "701667486/train1": 28.0, + "701667486/train2": 28.0, + "747260437/train0": 28.0, + "747260437/train1": 28.0, + "747260437/train2": 28.0, + "750448296/train0": 28.0, + "750448296/train1": 28.0, + "750448296/train2": 28.0, + "767747424/train0": 22.75, + "767747424/train1": 28.0, + "767747424/train2": 28.0 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 27.927272727272726, + "low": 24.69563636363636, + "mean": 26.348969696969696, + "n": 5, + "num_samples": 1000 + }, + "mean": 26.348969696969696, + "per_task": { + "LiftPegUpright-v1": 27.933333333333334, + "PickCube-v1": 28.0, + "PullCube-v1": 24.16, + "PushCube-v1": 23.833333333333332, + "StackCube-v1": 27.818181818181817 + } + }, + "metric": "pairwise_causal_calibration_pairs", + "micro": { + "confidence": 0.95, + "high": 26.604166666666668, + "low": 24.104166666666668, + "mean": 25.430555555555557, + "n": 144, + "num_samples": 1000 + }, + "n": 144 + }, + "predicted_margin": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": -0.173344961671436, + "low": -0.24109860738595115, + "mean": -0.20713680477602725, + "n": 120, + "num_samples": 1000 + }, + "mean": -0.20713680477602725, + "per_seed": { + "1073161778/train0": -0.16932960943008368, + "1073161778/train1": -0.17924265036524914, + "1073161778/train2": -0.18893577023738622, + "1113371261/train0": 0.10543872296223779, + "1113371261/train1": 0.014134138393897249, + "1113371261/train2": 0.10321814561223094, + "1151410346/train0": -0.09583695341622606, + "1151410346/train1": -0.14552686699164114, + "1151410346/train2": -0.11137550392569598, + "1261930456/train0": -0.30952529615270524, + "1261930456/train1": -0.34495061079317235, + "1261930456/train2": -0.3104603102303952, + "1324133122/train0": 0.09214616076360858, + "1324133122/train1": 0.006220904061644952, + "1324133122/train2": 0.07453570390221465, + "1324642411/train0": -0.5184929051461216, + "1324642411/train1": -0.38742954134254165, + "1324642411/train2": -0.17737669925866156, + "1375492786/train0": -0.2742629480034105, + "1375492786/train1": -0.19548314582621923, + "1375492786/train2": -0.2954350061136046, + "141984615/train0": -0.3913587564835576, + "141984615/train1": -0.3317204404518946, + "141984615/train2": -0.1614581645625799, + "1461069713/train0": -0.3069393465412038, + "1461069713/train1": -0.6098197544005363, + "1461069713/train2": -0.30833078189061486, + "1480856810/train0": -0.3484014667432773, + "1480856810/train1": -0.6088141163584672, + "1480856810/train2": -0.41400203151287207, + "1509980664/train0": -0.09419003293777646, + "1509980664/train1": -0.10500215845255088, + "1509980664/train2": -0.1890976986089807, + "1522272686/train0": 0.48462039184575295, + "1522272686/train1": 0.23140829517862893, + "1522272686/train2": 0.12607764557193257, + "1549324805/train0": -0.27807849592968054, + "1549324805/train1": -0.22126079522588438, + "1549324805/train2": -0.39631033691994494, + "1578990937/train0": -0.055414585551092216, + "1578990937/train1": -0.07562260772605807, + "1578990937/train2": -0.11122156610391924, + "1747178030/train0": -0.44234982955420926, + "1747178030/train1": -0.5995592153240683, + "1747178030/train2": -0.5527239930103893, + "1790188787/train0": -0.39804398287220066, + "1790188787/train1": -0.41064536165598464, + "1790188787/train2": -0.37508336216163984, + "1822210471/train0": -0.11426068280426432, + "1822210471/train1": -0.1660994968065328, + "1822210471/train2": -0.12061726254527255, + "1845489013/train0": -0.18030681051640662, + "1845489013/train1": -0.1558320961817238, + "1845489013/train2": -0.15437862688701337, + "1999022574/train0": -0.010543651527821427, + "1999022574/train1": -0.09841891022097132, + "1999022574/train2": -0.11507066220959525, + "2012211287/train0": -0.32231182251355883, + "2012211287/train1": -0.3136591221973332, + "2012211287/train2": -0.21985746212543253, + "2038594191/train0": 0.064572249902453, + "2038594191/train1": 0.0309942873094304, + "2038594191/train2": 0.09610840435854003, + "2054899126/train0": -0.43322840250453093, + "2054899126/train1": -0.4369772845393875, + "2054899126/train2": -0.4175765945025155, + "2061635888/train0": 0.07878378828125482, + "2061635888/train1": 0.004872853139381891, + "2061635888/train2": 0.061471356218488246, + "211245842/train0": -0.127498319380692, + "211245842/train1": -0.25192844397522873, + "211245842/train2": -0.19743010297696217, + "278008524/train0": -0.2659005496346014, + "278008524/train1": -0.2704948697931074, + "278008524/train2": -0.313945285637824, + "307068879/train0": -0.05797252706199239, + "307068879/train1": 0.018932161107400383, + "307068879/train2": -0.0032027848618630603, + "412102563/train0": -0.14882605428987036, + "412102563/train1": -0.26977222222104225, + "412102563/train2": -0.24386219595665565, + "440577711/train0": -0.41396662095926473, + "440577711/train1": -0.549285469979439, + "440577711/train2": -0.3642232105295439, + "441046872/train0": -0.06101528493441861, + "441046872/train1": -0.018316727382166484, + "441046872/train2": -0.037788380273409077, + "509430392/train0": -0.426096250608377, + "509430392/train1": -0.42675791108199884, + "509430392/train2": -0.32887303074558316, + "585904130/train0": -0.1685022894131509, + "585904130/train1": -0.1881172524610201, + "585904130/train2": -0.014285494888099215, + "608016290/train0": -0.2021012748656509, + "608016290/train1": -0.22391340049944683, + "608016290/train2": -0.1967433702484556, + "626486813/train0": -0.39959073191011746, + "626486813/train1": -0.403655648428104, + "626486813/train2": -0.26702454174718965, + "636680137/train0": -0.4612658703790792, + "636680137/train1": -0.4800313079785443, + "636680137/train2": -0.42177891564909986, + "655884605/train0": 0.06928250847505647, + "655884605/train1": 0.07124935757530594, + "655884605/train2": 0.0047442677921717205, + "672328262/train0": -0.008075281416847935, + "672328262/train1": -0.17563335554996382, + "672328262/train2": -0.014732418206182048, + "701667486/train0": -0.23997313317310848, + "701667486/train1": -0.2717920123841143, + "701667486/train2": -0.34033401997115287, + "747260437/train0": -0.3043353554007297, + "747260437/train1": -0.41501670336545593, + "747260437/train2": -0.41403821961369863, + "750448296/train0": -0.18584857501276866, + "750448296/train1": -0.11894762128301756, + "750448296/train2": -0.1279298521209656, + "767747424/train0": -0.3105746772245132, + "767747424/train1": -0.3587369662558639, + "767747424/train2": -0.35684179754966683 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.096719610649919, + "low": -0.3087825951404188, + "mean": -0.12359432046513516, + "n": 5, + "num_samples": 1000 + }, + "mean": -0.12359432046513516, + "per_task": { + "LiftPegUpright-v1": -0.14503053505926972, + "PickCube-v1": 0.28070211086543817, + "PullCube-v1": -0.27933737647327517, + "PushCube-v1": -0.3951348987851248, + "StackCube-v1": -0.07917090287344432 + } + }, + "metric": "predicted_margin", + "micro": { + "confidence": 0.95, + "high": -0.19492162677922983, + "low": -0.2523546562541373, + "mean": -0.22228279792877975, + "n": 144, + "num_samples": 1000 + }, + "n": 144 + }, + "proposal_oracle_success": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.29583333333333334, + "low": 0.15416666666666667, + "mean": 0.22291666666666668, + "n": 120, + "num_samples": 1000 + }, + "mean": 0.22291666666666668, + "per_seed": { + "1073161778/train0": 0.0, + "1073161778/train1": 0.0, + "1073161778/train2": 1.0, + "1113371261/train0": 0.0, + "1113371261/train1": 0.0, + "1113371261/train2": 0.0, + "1151410346/train0": 0.0, + "1151410346/train1": 0.0, + "1151410346/train2": 0.0, + "1261930456/train0": 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": 1.0, + "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": 0.0, + "1509980664/train0": 1.0, + "1509980664/train1": 1.0, + "1509980664/train2": 0.0, + "1522272686/train0": 0.0, + "1522272686/train1": 0.0, + "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": 1.0, + "1747178030/train1": 1.0, + "1747178030/train2": 1.0, + "1790188787/train0": 1.0, + "1790188787/train1": 0.0, + "1790188787/train2": 1.0, + "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": 1.0, + "1999022574/train2": 0.0, + "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": 0.5, + "2054899126/train1": 0.5, + "2054899126/train2": 0.5, + "2061635888/train0": 0.0, + "2061635888/train1": 0.0, + "2061635888/train2": 0.0, + "211245842/train0": 0.0, + "211245842/train1": 0.0, + "211245842/train2": 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": 0.0, + "440577711/train0": 0.0, + "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": 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": 0.0, + "672328262/train1": 0.0, + "672328262/train2": 0.0, + "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": 0.0, + "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": 0.40606060606060607, + "low": 0.15903030303030302, + "mean": 0.2754747474747475, + "n": 5, + "num_samples": 1000 + }, + "mean": 0.2754747474747475, + "per_task": { + "LiftPegUpright-v1": 0.13333333333333333, + "PickCube-v1": 0.3333333333333333, + "PullCube-v1": 0.17333333333333334, + "PushCube-v1": 0.5555555555555556, + "StackCube-v1": 0.18181818181818182 + } + }, + "metric": "proposal_oracle_success", + "micro": { + "confidence": 0.95, + "high": 0.2916666666666667, + "low": 0.1597222222222222, + "mean": 0.2222222222222222, + "n": 144, + "num_samples": 1000 + }, + "n": 144 + }, + "proposal_oracle_utility": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.7715499536289523, + "low": 0.5113214392965246, + "mean": 0.6364790620334033, + "n": 120, + "num_samples": 1000 + }, + "mean": 0.6364790620334033, + "per_seed": { + "1073161778/train0": 0.7180102467536926, + "1073161778/train1": 0.7134289145469666, + "1073161778/train2": 2.0, + "1113371261/train0": 0.0740400105714798, + "1113371261/train1": 0.09622243046760559, + "1113371261/train2": 0.1003647893667221, + "1151410346/train0": 0.5684944987297058, + "1151410346/train1": 0.2439272701740265, + "1151410346/train2": 0.5630103945732117, + "1261930456/train0": 0.09678153693675995, + "1261930456/train1": 0.09997129440307617, + "1261930456/train2": 0.08545441925525665, + "1324133122/train0": 0.0628519058227539, + "1324133122/train1": 0.0843733698129654, + "1324133122/train2": 0.09815146028995514, + "1324642411/train0": 0.21000581979751587, + "1324642411/train1": 2.0, + "1324642411/train2": 2.0, + "1375492786/train0": 2.0, + "1375492786/train1": 0.14743654429912567, + "1375492786/train2": 0.19452734291553497, + "141984615/train0": 0.023719489574432373, + "141984615/train1": 0.04397805407643318, + "141984615/train2": 0.04447716474533081, + "1461069713/train0": 0.05616961047053337, + "1461069713/train1": 0.05644957348704338, + "1461069713/train2": 0.11542954295873642, + "1480856810/train0": 0.14178013801574707, + "1480856810/train1": 2.0, + "1480856810/train2": 0.19622710347175598, + "1509980664/train0": 2.0, + "1509980664/train1": 2.0, + "1509980664/train2": 0.83889240026474, + "1522272686/train0": 0.5233199596405029, + "1522272686/train1": 0.5422300696372986, + "1522272686/train2": 2.0, + "1549324805/train0": 0.21791090071201324, + "1549324805/train1": 0.3088321089744568, + "1549324805/train2": 0.21403594315052032, + "1578990937/train0": 0.21425709128379822, + "1578990937/train1": 0.21763037145137787, + "1578990937/train2": 0.23259200155735016, + "1747178030/train0": 2.0, + "1747178030/train1": 2.0, + "1747178030/train2": 2.0, + "1790188787/train0": 2.0, + "1790188787/train1": 0.1862330436706543, + "1790188787/train2": 2.0, + "1822210471/train0": 0.03802649304270744, + "1822210471/train1": 0.07539409399032593, + "1822210471/train2": 0.058819372206926346, + "1845489013/train0": 0.0428844690322876, + "1845489013/train1": 0.09529292583465576, + "1845489013/train2": 0.0861951932311058, + "1999022574/train0": 0.5970457196235657, + "1999022574/train1": 2.0, + "1999022574/train2": 0.7970014810562134, + "2012211287/train0": 0.7277753949165344, + "2012211287/train1": 2.0, + "2012211287/train2": 0.7270426154136658, + "2038594191/train0": 0.15247176587581635, + "2038594191/train1": 0.1902807205915451, + "2038594191/train2": 0.24880830943584442, + "2054899126/train0": 1.0726270377635956, + "2054899126/train1": 1.0755530372262, + "2054899126/train2": 1.0683744624257088, + "2061635888/train0": 0.07642464339733124, + "2061635888/train1": 0.09046469628810883, + "2061635888/train2": 0.07902413606643677, + "211245842/train0": 0.5822281837463379, + "211245842/train1": 0.2016763538122177, + "211245842/train2": 0.2142850160598755, + "278008524/train0": 1.0722030997276306, + "278008524/train1": 1.1246033310890198, + "278008524/train2": 1.157032236456871, + "307068879/train0": 0.20280277729034424, + "307068879/train1": 0.20222051441669464, + "307068879/train2": 0.2035302072763443, + "412102563/train0": 2.0, + "412102563/train1": 2.0, + "412102563/train2": 0.8275061845779419, + "440577711/train0": 0.2057151049375534, + "440577711/train1": 2.0, + "440577711/train2": 2.0, + "441046872/train0": 0.203963503241539, + "441046872/train1": 0.20609354972839355, + "441046872/train2": 0.2042645812034607, + "509430392/train0": 2.0, + "509430392/train1": 2.0, + "509430392/train2": 2.0, + "585904130/train0": 0.16448359191417694, + "585904130/train1": 0.15698926150798798, + "585904130/train2": 0.2938849627971649, + "608016290/train0": 0.6124615669250488, + "608016290/train1": 0.6101007461547852, + "608016290/train2": 0.5705108642578125, + "626486813/train0": 0.05774742551147938, + "626486813/train1": 0.0619367565959692, + "626486813/train2": 0.08006523177027702, + "636680137/train0": 0.10778365284204483, + "636680137/train1": 0.11253269761800766, + "636680137/train2": 0.1247115358710289, + "655884605/train0": 0.15739890933036804, + "655884605/train1": 0.15038728713989258, + "655884605/train2": 0.17242512106895447, + "672328262/train0": 0.05979333445429802, + "672328262/train1": 0.053388457745313644, + "672328262/train2": 0.07216622680425644, + "701667486/train0": 0.7879213094711304, + "701667486/train1": 0.7947455048561096, + "701667486/train2": 2.0, + "747260437/train0": 0.19230616092681885, + "747260437/train1": 0.1342230886220932, + "747260437/train2": 0.18527252972126007, + "750448296/train0": 0.20943730572859445, + "750448296/train1": 0.22532296180725098, + "750448296/train2": 0.2481476068496704, + "767747424/train0": 0.5789013803005219, + "767747424/train1": 0.5825987188145518, + "767747424/train2": 0.5809952197596431 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 1.021850009759267, + "low": 0.5703265610120514, + "mean": 0.7913498574552512, + "n": 5, + "num_samples": 1000 + }, + "mean": 0.7913498574552512, + "per_task": { + "LiftPegUpright-v1": 0.6601470321416855, + "PickCube-v1": 1.021850009759267, + "PullCube-v1": 0.46170220340291657, + "PushCube-v1": 1.1790093589160178, + "StackCube-v1": 0.634040683056369 + } + }, + "metric": "proposal_oracle_utility", + "micro": { + "confidence": 0.95, + "high": 0.7532538059684966, + "low": 0.5049100180880891, + "mean": 0.6232009150553495, + "n": 144, + "num_samples": 1000 + }, + "n": 144 + }, + "selected_candidate_index": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 3.4729166666666673, + "low": 2.6888888888888887, + "mean": 3.1125, + "n": 120, + "num_samples": 1000 + }, + "mean": 3.1125, + "per_seed": { + "1073161778/train0": 3.0, + "1073161778/train1": 6.0, + "1073161778/train2": 5.0, + "1113371261/train0": 0.0, + "1113371261/train1": 4.0, + "1113371261/train2": 2.0, + "1151410346/train0": 2.0, + "1151410346/train1": 3.0, + "1151410346/train2": 3.0, + "1261930456/train0": 7.0, + "1261930456/train1": 4.0, + "1261930456/train2": 6.0, + "1324133122/train0": 4.0, + "1324133122/train1": 7.0, + "1324133122/train2": 4.0, + "1324642411/train0": 1.0, + "1324642411/train1": 0.0, + "1324642411/train2": 3.0, + "1375492786/train0": 1.0, + "1375492786/train1": 7.0, + "1375492786/train2": 2.0, + "141984615/train0": 6.0, + "141984615/train1": 5.0, + "141984615/train2": 1.0, + "1461069713/train0": 3.0, + "1461069713/train1": 1.0, + "1461069713/train2": 1.0, + "1480856810/train0": 5.0, + "1480856810/train1": 0.0, + "1480856810/train2": 4.0, + "1509980664/train0": 0.0, + "1509980664/train1": 1.0, + "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": 5.0, + "1578990937/train0": 4.0, + "1578990937/train1": 4.0, + "1578990937/train2": 5.0, + "1747178030/train0": 1.0, + "1747178030/train1": 4.0, + "1747178030/train2": 3.0, + "1790188787/train0": 0.0, + "1790188787/train1": 4.0, + "1790188787/train2": 1.0, + "1822210471/train0": 5.0, + "1822210471/train1": 5.0, + "1822210471/train2": 3.0, + "1845489013/train0": 0.0, + "1845489013/train1": 5.0, + "1845489013/train2": 3.0, + "1999022574/train0": 1.0, + "1999022574/train1": 0.0, + "1999022574/train2": 2.0, + "2012211287/train0": 2.0, + "2012211287/train1": 1.0, + "2012211287/train2": 4.0, + "2038594191/train0": 7.0, + "2038594191/train1": 4.0, + "2038594191/train2": 7.0, + "2054899126/train0": 4.5, + "2054899126/train1": 3.0, + "2054899126/train2": 0.0, + "2061635888/train0": 0.0, + "2061635888/train1": 4.0, + "2061635888/train2": 2.0, + "211245842/train0": 2.0, + "211245842/train1": 2.0, + "211245842/train2": 4.0, + "278008524/train0": 3.0, + "278008524/train1": 2.5, + "278008524/train2": 2.5, + "307068879/train0": 6.0, + "307068879/train1": 1.0, + "307068879/train2": 4.0, + "412102563/train0": 0.0, + "412102563/train1": 1.0, + "412102563/train2": 1.0, + "440577711/train0": 1.0, + "440577711/train1": 2.0, + "440577711/train2": 1.0, + "441046872/train0": 7.0, + "441046872/train1": 7.0, + "441046872/train2": 6.0, + "509430392/train0": 2.0, + "509430392/train1": 4.0, + "509430392/train2": 1.0, + "585904130/train0": 7.0, + "585904130/train1": 3.0, + "585904130/train2": 5.0, + "608016290/train0": 7.0, + "608016290/train1": 3.0, + "608016290/train2": 3.0, + "626486813/train0": 2.5, + "626486813/train1": 4.0, + "626486813/train2": 5.0, + "636680137/train0": 5.0, + "636680137/train1": 3.0, + "636680137/train2": 1.0, + "655884605/train0": 7.0, + "655884605/train1": 6.0, + "655884605/train2": 6.0, + "672328262/train0": 2.0, + "672328262/train1": 6.0, + "672328262/train2": 6.0, + "701667486/train0": 1.0, + "701667486/train1": 2.0, + "701667486/train2": 7.0, + "747260437/train0": 2.0, + "747260437/train1": 0.0, + "747260437/train2": 2.0, + "750448296/train0": 5.666666666666667, + "750448296/train1": 3.3333333333333335, + "750448296/train2": 3.0, + "767747424/train0": 2.25, + "767747424/train1": 2.75, + "767747424/train2": 3.5 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 3.703030303030303, + "low": 1.7644444444444445, + "mean": 2.7826262626262626, + "n": 5, + "num_samples": 1000 + }, + "mean": 2.7826262626262626, + "per_task": { + "LiftPegUpright-v1": 4.333333333333333, + "PickCube-v1": 1.0, + "PullCube-v1": 3.2666666666666666, + "PushCube-v1": 2.5555555555555554, + "StackCube-v1": 2.757575757575758 + } + }, + "metric": "selected_candidate_index", + "micro": { + "confidence": 0.95, + "high": 3.486111111111111, + "low": 2.7569444444444446, + "mean": 3.125, + "n": 144, + "num_samples": 1000 + }, + "n": 144 + }, + "selected_success": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.3694444444444444, + "low": 0.20277777777777778, + "mean": 0.28055555555555556, + "n": 120, + "num_samples": 1000 + }, + "mean": 0.28055555555555556, + "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": 1.0, + "1261930456/train1": 1.0, + "1261930456/train2": 1.0, + "1324133122/train0": 0.0, + "1324133122/train1": 0.0, + "1324133122/train2": 0.0, + "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": 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": 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": 1.0, + "1790188787/train2": 1.0, + "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": 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": 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": 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": 0.0, + "672328262/train1": 0.0, + "672328262/train2": 0.0, + "701667486/train0": 0.0, + "701667486/train1": 0.0, + "701667486/train2": 0.0, + "747260437/train0": 1.0, + "747260437/train1": 1.0, + "747260437/train2": 1.0, + "750448296/train0": 0.3333333333333333, + "750448296/train1": 0.3333333333333333, + "750448296/train2": 0.0, + "767747424/train0": 0.5, + "767747424/train1": 0.5, + "767747424/train2": 0.5 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.5693333333333334, + "low": 0.01818181818181818, + "mean": 0.2541818181818182, + "n": 5, + "num_samples": 1000 + }, + "mean": 0.2541818181818182, + "per_task": { + "LiftPegUpright-v1": 0.0, + "PickCube-v1": 0.0, + "PullCube-v1": 0.3466666666666667, + "PushCube-v1": 0.8333333333333334, + "StackCube-v1": 0.09090909090909091 + } + }, + "metric": "selected_success", + "micro": { + "confidence": 0.95, + "high": 0.3819444444444444, + "low": 0.2361111111111111, + "mean": 0.3055555555555556, + "n": 144, + "num_samples": 1000 + }, + "n": 144 + }, + "selected_success_gain_over_base": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": -0.0027777777777777775, + "low": -0.06111111111111111, + "mean": -0.02777777777777778, + "n": 120, + "num_samples": 1000 + }, + "mean": -0.02777777777777778, + "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": 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": 0.0, + "2054899126/train1": 0.0, + "2054899126/train2": 0.0, + "2061635888/train0": 0.0, + "2061635888/train1": 0.0, + "2061635888/train2": 0.0, + "211245842/train0": 0.0, + "211245842/train1": 0.0, + "211245842/train2": 0.0, + "278008524/train0": 0.0, + "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": 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": 0.0, + "672328262/train1": 0.0, + "672328262/train2": 0.0, + "701667486/train0": 0.0, + "701667486/train1": 0.0, + "701667486/train2": 0.0, + "747260437/train0": 0.0, + "747260437/train1": 0.0, + "747260437/train2": 0.0, + "750448296/train0": 0.0, + "750448296/train1": 0.0, + "750448296/train2": -0.3333333333333333, + "767747424/train0": 0.0, + "767747424/train1": 0.0, + "767747424/train2": 0.0 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.0, + "low": -0.6026666666666667, + "mean": -0.2026666666666667, + "n": 5, + "num_samples": 1000 + }, + "mean": -0.2026666666666667, + "per_task": { + "LiftPegUpright-v1": 0.0, + "PickCube-v1": -1.0, + "PullCube-v1": -0.013333333333333334, + "PushCube-v1": 0.0, + "StackCube-v1": 0.0 + } + }, + "metric": "selected_success_gain_over_base", + "micro": { + "confidence": 0.95, + "high": -0.006944444444444444, + "low": -0.05555555555555555, + "mean": -0.027777777777777776, + "n": 144, + "num_samples": 1000 + }, + "n": 144 + }, + "selected_utility": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.903949356648243, + "low": 0.6190276769610743, + "mean": 0.7537992654471761, + "n": 120, + "num_samples": 1000 + }, + "mean": 0.7537992654471761, + "per_seed": { + "1073161778/train0": 0.725517988204956, + "1073161778/train1": 0.725517988204956, + "1073161778/train2": 0.725517988204956, + "1113371261/train0": 0.0680118203163147, + "1113371261/train1": 0.08607026934623718, + "1113371261/train2": 0.09928342700004578, + "1151410346/train0": 0.2284735143184662, + "1151410346/train1": 0.2284735143184662, + "1151410346/train2": 0.2284735143184662, + "1261930456/train0": 2.0, + "1261930456/train1": 2.0, + "1261930456/train2": 2.0, + "1324133122/train0": 0.05331718921661377, + "1324133122/train1": 0.06854432821273804, + "1324133122/train2": 0.09103855490684509, + "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": 0.16783273220062256, + "141984615/train1": 0.16783273220062256, + "141984615/train2": 0.16783273220062256, + "1461069713/train0": 0.22852759063243866, + "1461069713/train1": 0.22852759063243866, + "1461069713/train2": 0.22852759063243866, + "1480856810/train0": 2.0, + "1480856810/train1": 2.0, + "1480856810/train2": 2.0, + "1509980664/train0": 0.8320798277854919, + "1509980664/train1": 0.8320798277854919, + "1509980664/train2": 0.8320798277854919, + "1522272686/train0": 0.13452494144439697, + "1522272686/train1": 0.09935368597507477, + "1522272686/train2": 0.5520380139350891, + "1549324805/train0": 0.17249532043933868, + "1549324805/train1": 0.17249532043933868, + "1549324805/train2": 0.17249532043933868, + "1578990937/train0": 0.21093030273914337, + "1578990937/train1": 0.21256697177886963, + "1578990937/train2": 0.21426597237586975, + "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": 0.09442546218633652, + "1822210471/train1": 0.09442546218633652, + "1822210471/train2": 0.09442546218633652, + "1845489013/train0": 0.11255800724029541, + "1845489013/train1": 0.11255800724029541, + "1845489013/train2": 0.11255800724029541, + "1999022574/train0": 0.5970457196235657, + "1999022574/train1": 0.8279688954353333, + "1999022574/train2": 0.8279688954353333, + "2012211287/train0": 0.7282466888427734, + "2012211287/train1": 0.7282466888427734, + "2012211287/train2": 0.7282466888427734, + "2038594191/train0": 0.15247176587581635, + "2038594191/train1": 0.14092111587524414, + "2038594191/train2": 0.23134692013263702, + "2054899126/train0": 2.0, + "2054899126/train1": 2.0, + "2054899126/train2": 2.0, + "2061635888/train0": 0.07517768442630768, + "2061635888/train1": 0.08908586204051971, + "2061635888/train2": 0.07194386422634125, + "211245842/train0": 0.18430444598197937, + "211245842/train1": 0.18430444598197937, + "211245842/train2": 0.18430444598197937, + "278008524/train0": 1.1250072121620178, + "278008524/train1": 1.1250072121620178, + "278008524/train2": 1.1250072121620178, + "307068879/train0": 0.19984664022922516, + "307068879/train1": 0.20222051441669464, + "307068879/train2": 0.2035302072763443, + "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": 0.203963503241539, + "441046872/train1": 0.20609354972839355, + "441046872/train2": 0.2042645812034607, + "509430392/train0": 0.04073460400104523, + "509430392/train1": 0.04073460400104523, + "509430392/train2": 0.04073460400104523, + "585904130/train0": 0.07103653997182846, + "585904130/train1": 0.07103653997182846, + "585904130/train2": 0.19301573932170868, + "608016290/train0": 0.525360107421875, + "608016290/train1": 0.525360107421875, + "608016290/train2": 0.525360107421875, + "626486813/train0": 0.04174089431762695, + "626486813/train1": 0.04174089431762695, + "626486813/train2": 0.04174089431762695, + "636680137/train0": 0.11957359313964844, + "636680137/train1": 0.11957359313964844, + "636680137/train2": 0.11957359313964844, + "655884605/train0": 0.13916455209255219, + "655884605/train1": 0.12485501170158386, + "655884605/train2": 0.1574591100215912, + "672328262/train0": 0.045009810477495193, + "672328262/train1": 0.10593423992395401, + "672328262/train2": 0.05270739272236824, + "701667486/train0": 0.7912740707397461, + "701667486/train1": 0.7912740707397461, + "701667486/train2": 0.7912740707397461, + "747260437/train0": 2.0, + "747260437/train1": 2.0, + "747260437/train2": 2.0, + "750448296/train0": 0.8105894525845846, + "750448296/train1": 0.7964938481648763, + "750448296/train2": 0.1862319732705752, + "767747424/train0": 1.0747087560594082, + "767747424/train1": 1.0747087560594082, + "767747424/train2": 1.0747087560594082 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 1.2138723237315812, + "low": 0.39336576275753254, + "mean": 0.7376363160483764, + "n": 5, + "num_samples": 1000 + }, + "mean": 0.7376363160483764, + "per_task": { + "LiftPegUpright-v1": 0.4771528899669647, + "PickCube-v1": 0.26197221378485364, + "PullCube-v1": 0.7870221028725306, + "PushCube-v1": 1.6734557673335075, + "StackCube-v1": 0.48857860628402594 + } + }, + "metric": "selected_utility", + "micro": { + "confidence": 0.95, + "high": 0.9183495887555182, + "low": 0.6529841326798002, + "mean": 0.7862164272616307, + "n": 144, + "num_samples": 1000 + }, + "n": 144 + }, + "selected_utility_gain_over_base": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": -0.007052166615095403, + "low": -0.10630760129748117, + "mean": -0.04900421073867215, + "n": 120, + "num_samples": 1000 + }, + "mean": -0.04900421073867215, + "per_seed": { + "1073161778/train0": 0.0, + "1073161778/train1": 0.0, + "1073161778/train2": 0.0, + "1113371261/train0": -0.043087467551231384, + "1113371261/train1": -0.0250290185213089, + "1113371261/train2": -0.011815860867500305, + "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": -0.030909866094589233, + "1324133122/train1": -0.015682727098464966, + "1324133122/train2": 0.00681149959564209, + "1324642411/train0": 0.0, + "1324642411/train1": 0.0, + "1324642411/train2": 0.0, + "1375492786/train0": 0.0, + "1375492786/train1": 0.0, + "1375492786/train2": 0.0, + "141984615/train0": 0.0, + "141984615/train1": 0.0, + "141984615/train2": 0.0, + "1461069713/train0": 0.0, + "1461069713/train1": 0.0, + "1461069713/train2": 0.0, + "1480856810/train0": 0.0, + "1480856810/train1": 0.0, + "1480856810/train2": 0.0, + "1509980664/train0": 0.0, + "1509980664/train1": 0.0, + "1509980664/train2": 0.0, + "1522272686/train0": -1.865475058555603, + "1522272686/train1": -1.9006463140249252, + "1522272686/train2": -1.4479619860649109, + "1549324805/train0": 0.0, + "1549324805/train1": 0.0, + "1549324805/train2": 0.0, + "1578990937/train0": -0.0033356696367263794, + "1578990937/train1": -0.001699000597000122, + "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": -0.23092317581176758, + "1999022574/train1": 0.0, + "1999022574/train2": 0.0, + "2012211287/train0": 0.0, + "2012211287/train1": 0.0, + "2012211287/train2": 0.0, + "2038594191/train0": 0.0574311688542366, + "2038594191/train1": 0.0458805188536644, + "2038594191/train2": 0.13630632311105728, + "2054899126/train0": 0.0, + "2054899126/train1": 0.0, + "2054899126/train2": 0.0, + "2061635888/train0": -0.011029243469238281, + "2061635888/train1": 0.002878934144973755, + "2061635888/train2": -0.014263063669204712, + "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": -0.009305715560913086, + "307068879/train1": -0.0069318413734436035, + "307068879/train2": -0.005622148513793945, + "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": -0.005804136395454407, + "441046872/train1": -0.0036740899085998535, + "441046872/train2": -0.005503058433532715, + "509430392/train0": 0.0, + "509430392/train1": 0.0, + "509430392/train2": 0.0, + "585904130/train0": 0.0, + "585904130/train1": 0.0, + "585904130/train2": 0.12197919934988022, + "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": -0.0049108415842056274, + "655884605/train1": -0.01922038197517395, + "655884605/train2": 0.013383716344833374, + "672328262/train0": -0.060924429446458817, + "672328262/train1": 0.0, + "672328262/train2": -0.05322684720158577, + "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": 0.0500861257314682, + "750448296/train1": 0.03599052131175995, + "750448296/train2": -0.5742713535825411, + "767747424/train0": 0.0, + "767747424/train1": 0.0, + "767747424/train2": 0.0 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": -0.0038301981940413966, + "low": -1.0465535867611568, + "mean": -0.35369903086171, + "n": 5, + "num_samples": 1000 + }, + "mean": -0.35369903086171, + "per_task": { + "LiftPegUpright-v1": -0.002456066012382507, + "PickCube-v1": -1.7380277862151463, + "PullCube-v1": -0.016228509147961933, + "PushCube-v1": 0.0, + "StackCube-v1": -0.011782792933059462 + } + }, + "metric": "selected_utility_gain_over_base", + "micro": { + "confidence": 0.95, + "high": -0.0061558029976569945, + "low": -0.09627005765731964, + "mean": -0.047617324317495026, + "n": 144, + "num_samples": 1000 + }, + "n": 144 + }, + "selector_regret": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.22856539063569573, + "low": 0.08062908659275209, + "mean": 0.1484731165174809, + "n": 120, + "num_samples": 1000 + }, + "mean": 0.1484731165174809, + "per_seed": { + "1073161778/train0": 0.0, + "1073161778/train1": 0.0, + "1073161778/train2": 1.274482011795044, + "1113371261/train0": 0.0060281902551651, + "1113371261/train1": 0.010152161121368408, + "1113371261/train2": 0.0010813623666763306, + "1151410346/train0": 0.3400209844112396, + "1151410346/train1": 0.015453755855560303, + "1151410346/train2": 0.3345368802547455, + "1261930456/train0": 0.0, + "1261930456/train1": 0.0, + "1261930456/train2": 0.0, + "1324133122/train0": 0.009534716606140137, + "1324133122/train1": 0.015829041600227356, + "1324133122/train2": 0.007112905383110046, + "1324642411/train0": 0.0, + "1324642411/train1": 0.0, + "1324642411/train2": 0.0, + "1375492786/train0": 0.0, + "1375492786/train1": 0.0, + "1375492786/train2": 0.0, + "141984615/train0": 0.0, + "141984615/train1": 0.0, + "141984615/train2": 0.0, + "1461069713/train0": 0.0, + "1461069713/train1": 0.0, + "1461069713/train2": 0.0, + "1480856810/train0": 0.0, + "1480856810/train1": 0.0, + "1480856810/train2": 0.0, + "1509980664/train0": 1.167920172214508, + "1509980664/train1": 1.167920172214508, + "1509980664/train2": 0.006812572479248047, + "1522272686/train0": 0.38879501819610596, + "1522272686/train1": 0.4428763836622238, + "1522272686/train2": 1.4479619860649109, + "1549324805/train0": 0.04541558027267456, + "1549324805/train1": 0.1363367885351181, + "1549324805/train2": 0.04154062271118164, + "1578990937/train0": 0.003326788544654846, + "1578990937/train1": 0.00506339967250824, + "1578990937/train2": 0.018326029181480408, + "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": 1.1720311045646667, + "1999022574/train2": 0.0, + "2012211287/train0": 0.0, + "2012211287/train1": 1.2717533111572266, + "2012211287/train2": 0.0, + "2038594191/train0": 0.0, + "2038594191/train1": 0.049359604716300964, + "2038594191/train2": 0.017461389303207397, + "2054899126/train0": 0.0, + "2054899126/train1": 0.0, + "2054899126/train2": 0.0, + "2061635888/train0": 0.0012469589710235596, + "2061635888/train1": 0.0013788342475891113, + "2061635888/train2": 0.00708027184009552, + "211245842/train0": 0.3979237377643585, + "211245842/train1": 0.017371907830238342, + "211245842/train2": 0.029980570077896118, + "278008524/train0": 0.0, + "278008524/train1": 0.0, + "278008524/train2": 0.03202502429485321, + "307068879/train0": 0.0029561370611190796, + "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": 0.0, + "441046872/train1": 0.0, + "441046872/train2": 0.0, + "509430392/train0": 1.9592653959989548, + "509430392/train1": 1.9592653959989548, + "509430392/train2": 1.9592653959989548, + "585904130/train0": 0.09344705194234848, + "585904130/train1": 0.08595272153615952, + "585904130/train2": 0.10086922347545624, + "608016290/train0": 0.08710145950317383, + "608016290/train1": 0.08474063873291016, + "608016290/train2": 0.0451507568359375, + "626486813/train0": 0.016006531193852425, + "626486813/train1": 0.020195862278342247, + "626486813/train2": 0.03832433745265007, + "636680137/train0": 0.0, + "636680137/train1": 0.0, + "636680137/train2": 0.005137942731380463, + "655884605/train0": 0.018234357237815857, + "655884605/train1": 0.025532275438308716, + "655884605/train2": 0.014966011047363281, + "672328262/train0": 0.014783523976802826, + "672328262/train1": 0.0, + "672328262/train2": 0.0194588340818882, + "701667486/train0": 0.0, + "701667486/train1": 0.0034714341163635254, + "701667486/train2": 1.208725929260254, + "747260437/train0": 0.0, + "747260437/train1": 0.0, + "747260437/train2": 0.0, + "750448296/train0": 0.013355518380800882, + "750448296/train1": 0.027911449472109478, + "750448296/train2": 0.07596839219331741, + "767747424/train0": 0.019455675035715103, + "767747424/train1": 0.019207116216421127, + "767747424/train2": 0.013884376734495163 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.5296915926535923, + "low": 0.11279337844252586, + "mean": 0.2942660002609094, + "n": 5, + "num_samples": 1000 + }, + "mean": 0.2942660002609094, + "per_task": { + "LiftPegUpright-v1": 0.18441228767236073, + "PickCube-v1": 0.7598777959744135, + "PullCube-v1": 0.01846348727742831, + "PushCube-v1": 0.32654423266649246, + "StackCube-v1": 0.18203219771385193 + } + }, + "metric": "selector_regret", + "micro": { + "confidence": 0.95, + "high": 0.19345460319891572, + "low": 0.070638673638718, + "mean": 0.12719054139840105, + "n": 144, + "num_samples": 1000 + }, + "n": 144 + }, + "success_selector_gap": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.14166666666666666, + "low": 0.041666666666666664, + "mean": 0.08333333333333333, + "n": 120, + "num_samples": 1000 + }, + "mean": 0.08333333333333333, + "per_seed": { + "1073161778/train0": 0.0, + "1073161778/train1": 0.0, + "1073161778/train2": 1.0, + "1113371261/train0": 0.0, + "1113371261/train1": 0.0, + "1113371261/train2": 0.0, + "1151410346/train0": 0.0, + "1151410346/train1": 0.0, + "1151410346/train2": 0.0, + "1261930456/train0": 0.0, + "1261930456/train1": 0.0, + "1261930456/train2": 0.0, + "1324133122/train0": 0.0, + "1324133122/train1": 0.0, + "1324133122/train2": 0.0, + "1324642411/train0": 0.0, + "1324642411/train1": 0.0, + "1324642411/train2": 0.0, + "1375492786/train0": 0.0, + "1375492786/train1": 0.0, + "1375492786/train2": 0.0, + "141984615/train0": 0.0, + "141984615/train1": 0.0, + "141984615/train2": 0.0, + "1461069713/train0": 0.0, + "1461069713/train1": 0.0, + "1461069713/train2": 0.0, + "1480856810/train0": 0.0, + "1480856810/train1": 0.0, + "1480856810/train2": 0.0, + "1509980664/train0": 1.0, + "1509980664/train1": 1.0, + "1509980664/train2": 0.0, + "1522272686/train0": 0.0, + "1522272686/train1": 0.0, + "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": 1.0, + "1999022574/train2": 0.0, + "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": 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": 0.0, + "440577711/train1": 0.0, + "440577711/train2": 0.0, + "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": 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": 0.0, + "672328262/train1": 0.0, + "672328262/train2": 0.0, + "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": 0.0, + "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": 0.2533333333333333, + "low": 0.05757575757575757, + "mean": 0.1509090909090909, + "n": 5, + "num_samples": 1000 + }, + "mean": 0.1509090909090909, + "per_task": { + "LiftPegUpright-v1": 0.13333333333333333, + "PickCube-v1": 0.3333333333333333, + "PullCube-v1": 0.0, + "PushCube-v1": 0.16666666666666666, + "StackCube-v1": 0.12121212121212122 + } + }, + "metric": "success_selector_gap", + "micro": { + "confidence": 0.95, + "high": 0.1111111111111111, + "low": 0.027777777777777776, + "mean": 0.06944444444444445, + "n": 144, + "num_samples": 1000 + }, + "n": 144 + }, + "success_support_gap": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 0.6597222222222222, + "low": 0.4847222222222222, + "mean": 0.5708333333333334, + "n": 120, + "num_samples": 1000 + }, + "mean": 0.5708333333333334, + "per_seed": { + "1073161778/train0": 1.0, + "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": 1.0, + "1324642411/train1": 0.0, + "1324642411/train2": 0.0, + "1375492786/train0": 0.0, + "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": 1.0, + "1522272686/train0": 1.0, + "1522272686/train1": 1.0, + "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": 0.0, + "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": 0.0, + "1999022574/train2": 1.0, + "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": 0.5, + "2054899126/train1": 0.5, + "2054899126/train2": 0.5, + "2061635888/train0": 1.0, + "2061635888/train1": 1.0, + "2061635888/train2": 1.0, + "211245842/train0": 1.0, + "211245842/train1": 1.0, + "211245842/train2": 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": 1.0, + "440577711/train0": 1.0, + "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": 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": 0.0, + "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": 0.0, + "767747424/train1": 0.0, + "767747424/train2": 0.0 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 0.8072727272727273, + "low": 0.40533333333333327, + "mean": 0.6063030303030303, + "n": 5, + "num_samples": 1000 + }, + "mean": 0.6063030303030303, + "per_task": { + "LiftPegUpright-v1": 0.8666666666666667, + "PickCube-v1": 0.6666666666666666, + "PullCube-v1": 0.3466666666666667, + "PushCube-v1": 0.3333333333333333, + "StackCube-v1": 0.8181818181818182 + } + }, + "metric": "success_support_gap", + "micro": { + "confidence": 0.95, + "high": 0.5972222222222222, + "low": 0.4375, + "mean": 0.5138888888888888, + "n": 144, + "num_samples": 1000 + }, + "n": 144 + }, + "support_gap": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": 1.1496533814558965, + "low": 0.8442967605000982, + "mean": 0.9982878092624661, + "n": 120, + "num_samples": 1000 + }, + "mean": 0.9982878092624661, + "per_seed": { + "1073161778/train0": 1.2819897532463074, + "1073161778/train1": 1.2865710854530334, + "1073161778/train2": 0.0, + "1113371261/train0": 1.9259599894285202, + "1113371261/train1": 1.9037775695323944, + "1113371261/train2": 1.899635210633278, + "1151410346/train0": 1.4315055012702942, + "1151410346/train1": 1.7560727298259735, + "1151410346/train2": 1.4369896054267883, + "1261930456/train0": 1.90321846306324, + "1261930456/train1": 1.9000287055969238, + "1261930456/train2": 1.9145455807447433, + "1324133122/train0": 1.937148094177246, + "1324133122/train1": 1.9156266301870346, + "1324133122/train2": 1.9018485397100449, + "1324642411/train0": 1.7899941802024841, + "1324642411/train1": 0.0, + "1324642411/train2": 0.0, + "1375492786/train0": 0.0, + "1375492786/train1": 1.8525634557008743, + "1375492786/train2": 1.805472657084465, + "141984615/train0": 0.2620577812194824, + "141984615/train1": 0.2417992167174816, + "141984615/train2": 0.24130010604858398, + "1461069713/train0": 0.10317284241318703, + "1461069713/train1": 0.10289287939667702, + "1461069713/train2": 0.04391290992498398, + "1480856810/train0": 0.06770022213459015, + "1480856810/train1": 0.0, + "1480856810/train2": 0.013253256678581238, + "1509980664/train0": 0.0, + "1509980664/train1": 0.0, + "1509980664/train2": 1.16110759973526, + "1522272686/train0": 1.476680040359497, + "1522272686/train1": 1.4577699303627014, + "1522272686/train2": 0.0, + "1549324805/train0": 0.07954518496990204, + "1549324805/train1": 0.0, + "1549324805/train2": 0.08342014253139496, + "1578990937/train0": 1.7857429087162018, + "1578990937/train1": 1.7823696285486221, + "1578990937/train2": 1.7674079984426498, + "1747178030/train0": 0.0, + "1747178030/train1": 0.0, + "1747178030/train2": 0.0, + "1790188787/train0": 0.0, + "1790188787/train1": 1.8137669563293457, + "1790188787/train2": 0.0, + "1822210471/train0": 1.9619735069572926, + "1822210471/train1": 1.924605906009674, + "1822210471/train2": 1.9411806277930737, + "1845489013/train0": 1.9571155309677124, + "1845489013/train1": 1.9047070741653442, + "1845489013/train2": 1.9138048067688942, + "1999022574/train0": 1.4029542803764343, + "1999022574/train1": 0.0, + "1999022574/train2": 1.2029985189437866, + "2012211287/train0": 1.2722246050834656, + "2012211287/train1": 0.0, + "2012211287/train2": 1.2729573845863342, + "2038594191/train0": 1.8475282341241837, + "2038594191/train1": 1.809719279408455, + "2038594191/train2": 1.7511916905641556, + "2054899126/train0": 0.9273729622364044, + "2054899126/train1": 0.9244469627737999, + "2054899126/train2": 0.9316255375742912, + "2061635888/train0": 1.9235753566026688, + "2061635888/train1": 1.9095353037118912, + "2061635888/train2": 1.9209758639335632, + "211245842/train0": 1.417771816253662, + "211245842/train1": 1.7983236461877823, + "211245842/train2": 1.7857149839401245, + "278008524/train0": 0.04170554131269455, + "278008524/train1": 0.0, + "278008524/train2": 0.0, + "307068879/train0": 1.7971972227096558, + "307068879/train1": 1.7977794855833054, + "307068879/train2": 1.7964697927236557, + "412102563/train0": 0.0, + "412102563/train1": 0.0, + "412102563/train2": 1.172493815422058, + "440577711/train0": 1.7942848950624466, + "440577711/train1": 0.0, + "440577711/train2": 0.0, + "441046872/train0": 1.796036496758461, + "441046872/train1": 1.7939064502716064, + "441046872/train2": 1.7957354187965393, + "509430392/train0": 0.0, + "509430392/train1": 0.0, + "509430392/train2": 0.0, + "585904130/train0": 1.835516408085823, + "585904130/train1": 1.843010738492012, + "585904130/train2": 1.706115037202835, + "608016290/train0": 1.3875384330749512, + "608016290/train1": 1.3898992538452148, + "608016290/train2": 1.4294891357421875, + "626486813/train0": 0.1078349556773901, + "626486813/train1": 0.10364562459290028, + "626486813/train2": 0.08551714941859245, + "636680137/train0": 0.06278067082166672, + "636680137/train1": 0.05803162604570389, + "636680137/train2": 0.04585278779268265, + "655884605/train0": 1.842601090669632, + "655884605/train1": 1.8496127128601074, + "655884605/train2": 1.8275748789310455, + "672328262/train0": 1.940206665545702, + "672328262/train1": 1.9466115422546864, + "672328262/train2": 1.9278337731957436, + "701667486/train0": 1.2120786905288696, + "701667486/train1": 1.2052544951438904, + "701667486/train2": 0.0, + "747260437/train0": 0.08635708689689636, + "747260437/train1": 0.144440159201622, + "747260437/train2": 0.09339071810245514, + "750448296/train0": 1.1985591898361843, + "750448296/train1": 1.1875135501225789, + "750448296/train2": 1.1598488887151082, + "767747424/train0": 0.10475751012563705, + "767747424/train1": 0.1008116127923131, + "767747424/train2": 0.09709237236529589 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": 1.2779549119689249, + "low": 0.7197053071856498, + "mean": 0.9988301095772872, + "n": 5, + "num_samples": 1000 + }, + "mean": 0.9988301095772872, + "per_task": { + "LiftPegUpright-v1": 1.3398529678583144, + "PickCube-v1": 0.9781499902407328, + "PullCube-v1": 0.6881442584097386, + "PushCube-v1": 0.6220440144340197, + "StackCube-v1": 1.365959316943631 + } + }, + "metric": "support_gap", + "micro": { + "confidence": 0.95, + "high": 1.045588894850678, + "low": 0.7773104406013671, + "mean": 0.9091424554741631, + "n": 144, + "num_samples": 1000 + }, + "n": 144 + }, + "tau": { + "macro_by_seed": { + "ci": { + "confidence": 0.95, + "high": -0.08708576405217451, + "low": -0.08708576405217451, + "mean": -0.08708576405217451, + "n": 120, + "num_samples": 1000 + }, + "mean": -0.08708576405217451, + "per_seed": { + "1073161778/train0": -0.08708576405217451, + "1073161778/train1": -0.08708576405217451, + "1073161778/train2": -0.08708576405217451, + "1113371261/train0": -0.08708576405217451, + "1113371261/train1": -0.08708576405217451, + "1113371261/train2": -0.08708576405217451, + "1151410346/train0": -0.08708576405217451, + "1151410346/train1": -0.08708576405217451, + "1151410346/train2": -0.08708576405217451, + "1261930456/train0": -0.08708576405217451, + "1261930456/train1": -0.08708576405217451, + "1261930456/train2": -0.08708576405217451, + "1324133122/train0": -0.08708576405217451, + "1324133122/train1": -0.08708576405217451, + "1324133122/train2": -0.08708576405217451, + "1324642411/train0": -0.08708576405217451, + "1324642411/train1": -0.08708576405217451, + "1324642411/train2": -0.08708576405217451, + "1375492786/train0": -0.08708576405217451, + "1375492786/train1": -0.08708576405217451, + "1375492786/train2": -0.08708576405217451, + "141984615/train0": -0.08708576405217451, + "141984615/train1": -0.08708576405217451, + "141984615/train2": -0.08708576405217451, + "1461069713/train0": -0.08708576405217451, + "1461069713/train1": -0.08708576405217451, + "1461069713/train2": -0.08708576405217451, + "1480856810/train0": -0.08708576405217451, + "1480856810/train1": -0.08708576405217451, + "1480856810/train2": -0.08708576405217451, + "1509980664/train0": -0.08708576405217451, + "1509980664/train1": -0.08708576405217451, + "1509980664/train2": -0.08708576405217451, + "1522272686/train0": -0.08708576405217451, + "1522272686/train1": -0.08708576405217451, + "1522272686/train2": -0.08708576405217451, + "1549324805/train0": -0.08708576405217451, + "1549324805/train1": -0.08708576405217451, + "1549324805/train2": -0.08708576405217451, + "1578990937/train0": -0.08708576405217451, + "1578990937/train1": -0.08708576405217451, + "1578990937/train2": -0.08708576405217451, + "1747178030/train0": -0.08708576405217451, + "1747178030/train1": -0.08708576405217451, + "1747178030/train2": -0.08708576405217451, + "1790188787/train0": -0.08708576405217451, + "1790188787/train1": -0.08708576405217451, + "1790188787/train2": -0.08708576405217451, + "1822210471/train0": -0.08708576405217451, + "1822210471/train1": -0.08708576405217451, + "1822210471/train2": -0.08708576405217451, + "1845489013/train0": -0.08708576405217451, + "1845489013/train1": -0.08708576405217451, + "1845489013/train2": -0.08708576405217451, + "1999022574/train0": -0.08708576405217451, + "1999022574/train1": -0.08708576405217451, + "1999022574/train2": -0.08708576405217451, + "2012211287/train0": -0.08708576405217451, + "2012211287/train1": -0.08708576405217451, + "2012211287/train2": -0.08708576405217451, + "2038594191/train0": -0.08708576405217451, + "2038594191/train1": -0.08708576405217451, + "2038594191/train2": -0.08708576405217451, + "2054899126/train0": -0.08708576405217451, + "2054899126/train1": -0.08708576405217451, + "2054899126/train2": -0.08708576405217451, + "2061635888/train0": -0.08708576405217451, + "2061635888/train1": -0.08708576405217451, + "2061635888/train2": -0.08708576405217451, + "211245842/train0": -0.08708576405217451, + "211245842/train1": -0.08708576405217451, + "211245842/train2": -0.08708576405217451, + "278008524/train0": -0.08708576405217451, + "278008524/train1": -0.08708576405217451, + "278008524/train2": -0.08708576405217451, + "307068879/train0": -0.08708576405217451, + "307068879/train1": -0.08708576405217451, + "307068879/train2": -0.08708576405217451, + "412102563/train0": -0.08708576405217451, + "412102563/train1": -0.08708576405217451, + "412102563/train2": -0.08708576405217451, + "440577711/train0": -0.08708576405217451, + "440577711/train1": -0.08708576405217451, + "440577711/train2": -0.08708576405217451, + "441046872/train0": -0.08708576405217451, + "441046872/train1": -0.08708576405217451, + "441046872/train2": -0.08708576405217451, + "509430392/train0": -0.08708576405217451, + "509430392/train1": -0.08708576405217451, + "509430392/train2": -0.08708576405217451, + "585904130/train0": -0.08708576405217451, + "585904130/train1": -0.08708576405217451, + "585904130/train2": -0.08708576405217451, + "608016290/train0": -0.08708576405217451, + "608016290/train1": -0.08708576405217451, + "608016290/train2": -0.08708576405217451, + "626486813/train0": -0.08708576405217451, + "626486813/train1": -0.08708576405217451, + "626486813/train2": -0.08708576405217451, + "636680137/train0": -0.08708576405217451, + "636680137/train1": -0.08708576405217451, + "636680137/train2": -0.08708576405217451, + "655884605/train0": -0.08708576405217451, + "655884605/train1": -0.08708576405217451, + "655884605/train2": -0.08708576405217451, + "672328262/train0": -0.08708576405217451, + "672328262/train1": -0.08708576405217451, + "672328262/train2": -0.08708576405217451, + "701667486/train0": -0.08708576405217451, + "701667486/train1": -0.08708576405217451, + "701667486/train2": -0.08708576405217451, + "747260437/train0": -0.08708576405217451, + "747260437/train1": -0.08708576405217451, + "747260437/train2": -0.08708576405217451, + "750448296/train0": -0.08708576405217451, + "750448296/train1": -0.08708576405217451, + "750448296/train2": -0.08708576405217451, + "767747424/train0": -0.08708576405217451, + "767747424/train1": -0.08708576405217451, + "767747424/train2": -0.08708576405217451 + } + }, + "macro_by_task": { + "ci": { + "confidence": 0.95, + "high": -0.08708576405217451, + "low": -0.08708576405217451, + "mean": -0.08708576405217451, + "n": 5, + "num_samples": 1000 + }, + "mean": -0.08708576405217451, + "per_task": { + "LiftPegUpright-v1": -0.08708576405217451, + "PickCube-v1": -0.08708576405217451, + "PullCube-v1": -0.08708576405217451, + "PushCube-v1": -0.08708576405217451, + "StackCube-v1": -0.08708576405217451 + } + }, + "metric": "tau", + "micro": { + "confidence": 0.95, + "high": -0.08708576405217451, + "low": -0.08708576405217451, + "mean": -0.08708576405217451, + "n": 144, + "num_samples": 1000 + }, + "n": 144 + } + }, + "target": "utility_margin", + "tau": -0.08708576405217451, + "threshold_scope": "global", + "weights": [ + -0.36334951815701466, + 0.033145095968695394, + -0.0940629569733999, + 0.03459280547861965, + 0.007387327289540752, + -0.0067149633260008895, + 0.08952147646809869, + -0.06641047589384971, + 0.09165865900574419, + 0.0 + ] +} diff --git a/workspace/runs/ctt_base_context_obs_learned_dominance_basic_perdim_trainmax_train_to_test/metrics_by_seed.json b/workspace/runs/ctt_base_context_obs_learned_dominance_basic_perdim_trainmax_train_to_test/metrics_by_seed.json new file mode 100644 index 0000000000000000000000000000000000000000..e98fda8b92b138d70dbd2b8c13b588009bcd8138 --- /dev/null +++ b/workspace/runs/ctt_base_context_obs_learned_dominance_basic_perdim_trainmax_train_to_test/metrics_by_seed.json @@ -0,0 +1,3234 @@ +{ + "1073161778/train0": { + "base_success": 0.0, + "base_utility": 0.725517988204956, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_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": 0.5176761329820321, + "pairwise_causal_calibration_ece": 0.017676132982032078, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.16932960943008368, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.7180102467536926, + "selected_candidate_index": 3.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.725517988204956, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.2819897532463074, + "tau": -0.08708576405217451 + }, + "1073161778/train1": { + "base_success": 0.0, + "base_utility": 0.725517988204956, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_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": 0.5189946754984194, + "pairwise_causal_calibration_ece": 0.01671961021586632, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.17924265036524914, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.7134289145469666, + "selected_candidate_index": 6.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.725517988204956, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.2865710854530334, + "tau": -0.08708576405217451 + }, + "1073161778/train2": { + "base_success": 0.0, + "base_utility": 0.725517988204956, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_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": 0.5177030926793956, + "pairwise_causal_calibration_ece": 0.053417378393681325, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.18893577023738622, + "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": 0.725517988204956, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.274482011795044, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": -0.08708576405217451 + }, + "1113371261/train0": { + "base_success": 0.0, + "base_utility": 0.11109928786754608, + "coverage": 1.0, + "execute_generated": 1.0, + "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": 0.5435152760722883, + "pairwise_causal_calibration_ece": 0.06362758107056876, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.10543872296223779, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.0740400105714798, + "selected_candidate_index": 0.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.0680118203163147, + "selected_utility_gain_over_base": -0.043087467551231384, + "selector_regret": 0.0060281902551651, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9259599894285202, + "tau": -0.08708576405217451 + }, + "1113371261/train1": { + "base_success": 0.0, + "base_utility": 0.11109928786754608, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.39285714285714285, + "pairwise_causal_calibration_confidence": 0.5447882952488178, + "pairwise_causal_calibration_ece": 0.15193115239167487, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.014134138393897249, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.09622243046760559, + "selected_candidate_index": 4.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.08607026934623718, + "selected_utility_gain_over_base": -0.0250290185213089, + "selector_regret": 0.010152161121368408, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9037775695323944, + "tau": -0.08708576405217451 + }, + "1113371261/train2": { + "base_success": 0.0, + "base_utility": 0.11109928786754608, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.6785714285714286, + "pairwise_causal_calibration_confidence": 0.5452679026603489, + "pairwise_causal_calibration_ece": 0.13330352591107966, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.10321814561223094, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.1003647893667221, + "selected_candidate_index": 2.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.09928342700004578, + "selected_utility_gain_over_base": -0.011815860867500305, + "selector_regret": 0.0010813623666763306, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.899635210633278, + "tau": -0.08708576405217451 + }, + "1151410346/train0": { + "base_success": 0.0, + "base_utility": 0.2284735143184662, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.32142857142857145, + "pairwise_causal_calibration_confidence": 0.5208530758222207, + "pairwise_causal_calibration_ece": 0.19942450439364928, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.09583695341622606, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.5684944987297058, + "selected_candidate_index": 2.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.2284735143184662, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.3400209844112396, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.4315055012702942, + "tau": -0.08708576405217451 + }, + "1151410346/train1": { + "base_success": 0.0, + "base_utility": 0.2284735143184662, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_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": 0.5245737063651857, + "pairwise_causal_calibration_ece": 0.31028799207947144, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.14552686699164114, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.2439272701740265, + "selected_candidate_index": 3.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.2284735143184662, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.015453755855560303, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7560727298259735, + "tau": -0.08708576405217451 + }, + "1151410346/train2": { + "base_success": 0.0, + "base_utility": 0.2284735143184662, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_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": 0.5243490641793181, + "pairwise_causal_calibration_ece": 0.1314919213221753, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.11137550392569598, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.5630103945732117, + "selected_candidate_index": 3.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.2284735143184662, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.3345368802547455, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.4369896054267883, + "tau": -0.08708576405217451 + }, + "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": 0.5, + "pairwise_causal_calibration_confidence": 0.5197191753325973, + "pairwise_causal_calibration_ece": 0.01971917533259726, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.30952529615270524, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.09678153693675995, + "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": 1.0, + "support_gap": 1.90321846306324, + "tau": -0.08708576405217451 + }, + "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": 0.32142857142857145, + "pairwise_causal_calibration_confidence": 0.5188298291687808, + "pairwise_causal_calibration_ece": 0.1974012577402094, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.34495061079317235, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.09997129440307617, + "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": 1.0, + "support_gap": 1.9000287055969238, + "tau": -0.08708576405217451 + }, + "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": 0.5714285714285714, + "pairwise_causal_calibration_confidence": 0.5214267898957143, + "pairwise_causal_calibration_ece": 0.05000178153285706, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.3104603102303952, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.08545441925525665, + "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": 1.0, + "support_gap": 1.9145455807447433, + "tau": -0.08708576405217451 + }, + "1324133122/train0": { + "base_success": 0.0, + "base_utility": 0.084227055311203, + "coverage": 1.0, + "execute_generated": 1.0, + "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": 0.5433225220880885, + "pairwise_causal_calibration_ece": 0.0076082363738028125, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.09214616076360858, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.0628519058227539, + "selected_candidate_index": 4.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.05331718921661377, + "selected_utility_gain_over_base": -0.030909866094589233, + "selector_regret": 0.009534716606140137, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.937148094177246, + "tau": -0.08708576405217451 + }, + "1324133122/train1": { + "base_success": 0.0, + "base_utility": 0.084227055311203, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6071428571428571, + "pairwise_causal_calibration_confidence": 0.5410803984713379, + "pairwise_causal_calibration_ece": 0.06606245867151916, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.006220904061644952, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.0843733698129654, + "selected_candidate_index": 7.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.06854432821273804, + "selected_utility_gain_over_base": -0.015682727098464966, + "selector_regret": 0.015829041600227356, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9156266301870346, + "tau": -0.08708576405217451 + }, + "1324133122/train2": { + "base_success": 0.0, + "base_utility": 0.084227055311203, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "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": 0.5453211435652117, + "pairwise_causal_calibration_ece": 0.1332502850062169, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.07453570390221465, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.09815146028995514, + "selected_candidate_index": 4.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.09103855490684509, + "selected_utility_gain_over_base": 0.00681149959564209, + "selector_regret": 0.007112905383110046, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9018485397100449, + "tau": -0.08708576405217451 + }, + "1324642411/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": 0.39285714285714285, + "pairwise_causal_calibration_confidence": 0.5148670429643715, + "pairwise_causal_calibration_ece": 0.12200990010722862, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.5184929051461216, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.21000581979751587, + "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": 1.0, + "support_gap": 1.7899941802024841, + "tau": -0.08708576405217451 + }, + "1324642411/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": 0.48148148148148145, + "pairwise_causal_calibration_confidence": 0.5161761768637876, + "pairwise_causal_calibration_ece": 0.03469469538230613, + "pairwise_causal_calibration_pairs": 27.0, + "predicted_margin": -0.38742954134254165, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 0.0, + "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": -0.08708576405217451 + }, + "1324642411/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": 0.3333333333333333, + "pairwise_causal_calibration_confidence": 0.5164803540872519, + "pairwise_causal_calibration_ece": 0.1831470207539186, + "pairwise_causal_calibration_pairs": 18.0, + "predicted_margin": -0.17737669925866156, + "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": -0.08708576405217451 + }, + "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": 0.3076923076923077, + "pairwise_causal_calibration_confidence": 0.5079098226596691, + "pairwise_causal_calibration_ece": 0.2002175149673614, + "pairwise_causal_calibration_pairs": 13.0, + "predicted_margin": -0.2742629480034105, + "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": -0.08708576405217451 + }, + "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": 0.39285714285714285, + "pairwise_causal_calibration_confidence": 0.5110081951001124, + "pairwise_causal_calibration_ece": 0.11815105224296957, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.19548314582621923, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.14743654429912567, + "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": 1.0, + "support_gap": 1.8525634557008743, + "tau": -0.08708576405217451 + }, + "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": 0.6428571428571429, + "pairwise_causal_calibration_confidence": 0.5145465188512656, + "pairwise_causal_calibration_ece": 0.12831062400587734, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.2954350061136046, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.19452734291553497, + "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": 1.0, + "support_gap": 1.805472657084465, + "tau": -0.08708576405217451 + }, + "141984615/train0": { + "base_success": 0.0, + "base_utility": 0.16783273220062256, + "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": 0.7857142857142857, + "pairwise_causal_calibration_confidence": 0.5131339872831697, + "pairwise_causal_calibration_ece": 0.272580298431116, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.3913587564835576, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.023719489574432373, + "selected_candidate_index": 6.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.16783273220062256, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.2620577812194824, + "tau": -0.08708576405217451 + }, + "141984615/train1": { + "base_success": 0.0, + "base_utility": 0.16783273220062256, + "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": 0.2857142857142857, + "pairwise_causal_calibration_confidence": 0.5141218630904353, + "pairwise_causal_calibration_ece": 0.22840757737614958, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.3317204404518946, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.04397805407643318, + "selected_candidate_index": 5.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.16783273220062256, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.2417992167174816, + "tau": -0.08708576405217451 + }, + "141984615/train2": { + "base_success": 0.0, + "base_utility": 0.16783273220062256, + "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": 0.4642857142857143, + "pairwise_causal_calibration_confidence": 0.5067099785179769, + "pairwise_causal_calibration_ece": 0.04242426423226264, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.1614581645625799, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.04447716474533081, + "selected_candidate_index": 1.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.16783273220062256, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.24130010604858398, + "tau": -0.08708576405217451 + }, + "1461069713/train0": { + "base_success": 0.0, + "base_utility": 0.22852759063243866, + "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": 0.5714285714285714, + "pairwise_causal_calibration_confidence": 0.5096685000943126, + "pairwise_causal_calibration_ece": 0.06176007133425876, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.3069393465412038, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.05616961047053337, + "selected_candidate_index": 3.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.22852759063243866, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.10317284241318703, + "tau": -0.08708576405217451 + }, + "1461069713/train1": { + "base_success": 0.0, + "base_utility": 0.22852759063243866, + "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": 0.5357142857142857, + "pairwise_causal_calibration_confidence": 0.5043374545649476, + "pairwise_causal_calibration_ece": 0.03137683114933809, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.6098197544005363, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.05644957348704338, + "selected_candidate_index": 1.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.22852759063243866, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.10289287939667702, + "tau": -0.08708576405217451 + }, + "1461069713/train2": { + "base_success": 0.0, + "base_utility": 0.22852759063243866, + "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": 0.5, + "pairwise_causal_calibration_confidence": 0.5107514653860995, + "pairwise_causal_calibration_ece": 0.010751465386099501, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.30833078189061486, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.11542954295873642, + "selected_candidate_index": 1.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.22852759063243866, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.04391290992498398, + "tau": -0.08708576405217451 + }, + "1480856810/train0": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.20948036015033722, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.7142857142857143, + "pairwise_causal_calibration_confidence": 0.5090541462607746, + "pairwise_causal_calibration_ece": 0.2052315680249397, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.3484014667432773, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.14178013801574707, + "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": 0.06770022213459015, + "tau": -0.08708576405217451 + }, + "1480856810/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": 0.20948036015033722, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.32142857142857145, + "pairwise_causal_calibration_confidence": 0.5099948857012214, + "pairwise_causal_calibration_ece": 0.18856631427264997, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.6088141163584672, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 0.0, + "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": -0.08708576405217451 + }, + "1480856810/train2": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.20948036015033722, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5357142857142857, + "pairwise_causal_calibration_confidence": 0.5037911655250441, + "pairwise_causal_calibration_ece": 0.031923120189241594, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.41400203151287207, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.19622710347175598, + "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": 0.013253256678581238, + "tau": -0.08708576405217451 + }, + "1509980664/train0": { + "base_success": 0.0, + "base_utility": 0.8320798277854919, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "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": 0.5230050972773443, + "pairwise_causal_calibration_ece": 0.12300509727734432, + "pairwise_causal_calibration_pairs": 25.0, + "predicted_margin": -0.09419003293777646, + "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": 0.8320798277854919, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.167920172214508, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": -0.08708576405217451 + }, + "1509980664/train1": { + "base_success": 0.0, + "base_utility": 0.8320798277854919, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "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": 0.524259944555156, + "pairwise_causal_calibration_ece": 0.03574005544484404, + "pairwise_causal_calibration_pairs": 25.0, + "predicted_margin": -0.10500215845255088, + "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": 0.8320798277854919, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.167920172214508, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": -0.08708576405217451 + }, + "1509980664/train2": { + "base_success": 0.0, + "base_utility": 0.8320798277854919, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.42857142857142855, + "pairwise_causal_calibration_confidence": 0.5212588879913773, + "pairwise_causal_calibration_ece": 0.09268745941994877, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.1890976986089807, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.83889240026474, + "selected_candidate_index": 0.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.8320798277854919, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.006812572479248047, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.16110759973526, + "tau": -0.08708576405217451 + }, + "1522272686/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": 0.35714285714285715, + "pairwise_causal_calibration_confidence": 0.5554844314022709, + "pairwise_causal_calibration_ece": 0.1983415742594138, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.48462039184575295, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.5233199596405029, + "selected_candidate_index": 1.0, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.13452494144439697, + "selected_utility_gain_over_base": -1.865475058555603, + "selector_regret": 0.38879501819610596, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.476680040359497, + "tau": -0.08708576405217451 + }, + "1522272686/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": 0.07142857142857142, + "pairwise_causal_calibration_confidence": 0.5480190882988387, + "pairwise_causal_calibration_ece": 0.4765905168702673, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.23140829517862893, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.5422300696372986, + "selected_candidate_index": 1.0, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.09935368597507477, + "selected_utility_gain_over_base": -1.9006463140249252, + "selector_regret": 0.4428763836622238, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.4577699303627014, + "tau": -0.08708576405217451 + }, + "1522272686/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": 0.5357142857142857, + "pairwise_causal_calibration_confidence": 0.5381827836063698, + "pairwise_causal_calibration_ece": 0.0024684978920841427, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.12607764557193257, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 1.0, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.5520380139350891, + "selected_utility_gain_over_base": -1.4479619860649109, + "selector_regret": 1.4479619860649109, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": -0.08708576405217451 + }, + "1549324805/train0": { + "base_success": 0.0, + "base_utility": 0.17249532043933868, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2974560856819153, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.8571428571428571, + "pairwise_causal_calibration_confidence": 0.519176648228606, + "pairwise_causal_calibration_ece": 0.3379662089142511, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.27807849592968054, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.21791090071201324, + "selected_candidate_index": 0.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.17249532043933868, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.04541558027267456, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.07954518496990204, + "tau": -0.08708576405217451 + }, + "1549324805/train1": { + "base_success": 0.0, + "base_utility": 0.17249532043933868, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2974560856819153, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.7857142857142857, + "pairwise_causal_calibration_confidence": 0.5195775633028077, + "pairwise_causal_calibration_ece": 0.266136722411478, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.22126079522588438, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.3088321089744568, + "selected_candidate_index": 0.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.17249532043933868, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.1363367885351181, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": -0.08708576405217451 + }, + "1549324805/train2": { + "base_success": 0.0, + "base_utility": 0.17249532043933868, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2974560856819153, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.7857142857142857, + "pairwise_causal_calibration_confidence": 0.5151424475596346, + "pairwise_causal_calibration_ece": 0.27057183815465113, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.39631033691994494, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.21403594315052032, + "selected_candidate_index": 5.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.17249532043933868, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.04154062271118164, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.08342014253139496, + "tau": -0.08708576405217451 + }, + "1578990937/train0": { + "base_success": 0.0, + "base_utility": 0.21426597237586975, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.2857142857142857, + "pairwise_causal_calibration_confidence": 0.5228634819661582, + "pairwise_causal_calibration_ece": 0.23714919625187247, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.055414585551092216, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.21425709128379822, + "selected_candidate_index": 4.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.21093030273914337, + "selected_utility_gain_over_base": -0.0033356696367263794, + "selector_regret": 0.003326788544654846, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7857429087162018, + "tau": -0.08708576405217451 + }, + "1578990937/train1": { + "base_success": 0.0, + "base_utility": 0.21426597237586975, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.7142857142857143, + "pairwise_causal_calibration_confidence": 0.5249253193280673, + "pairwise_causal_calibration_ece": 0.189360394957647, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.07562260772605807, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.21763037145137787, + "selected_candidate_index": 4.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.21256697177886963, + "selected_utility_gain_over_base": -0.001699000597000122, + "selector_regret": 0.00506339967250824, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7823696285486221, + "tau": -0.08708576405217451 + }, + "1578990937/train2": { + "base_success": 0.0, + "base_utility": 0.21426597237586975, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_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": 0.5243615747594301, + "pairwise_causal_calibration_ece": 0.024361574759430127, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.11122156610391924, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.23259200155735016, + "selected_candidate_index": 5.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.21426597237586975, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.018326029181480408, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7674079984426498, + "tau": -0.08708576405217451 + }, + "1747178030/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_ece": 0.0, + "pairwise_causal_calibration_pairs": 0.0, + "predicted_margin": -0.44234982955420926, + "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": -0.08708576405217451 + }, + "1747178030/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_ece": 0.0, + "pairwise_causal_calibration_pairs": 0.0, + "predicted_margin": -0.5995592153240683, + "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": -0.08708576405217451 + }, + "1747178030/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_ece": 0.0, + "pairwise_causal_calibration_pairs": 0.0, + "predicted_margin": -0.5527239930103893, + "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": -0.08708576405217451 + }, + "1790188787/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": 0.8928571428571429, + "pairwise_causal_calibration_confidence": 0.5043835056714361, + "pairwise_causal_calibration_ece": 0.3884736371857068, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.39804398287220066, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 0.0, + "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": -0.08708576405217451 + }, + "1790188787/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": 0.5357142857142857, + "pairwise_causal_calibration_confidence": 0.5065734080456246, + "pairwise_causal_calibration_ece": 0.029140877668661136, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.41064536165598464, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.1862330436706543, + "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": 1.0, + "support_gap": 1.8137669563293457, + "tau": -0.08708576405217451 + }, + "1790188787/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": 0.8636363636363636, + "pairwise_causal_calibration_confidence": 0.5093146507839553, + "pairwise_causal_calibration_ece": 0.3543217128524083, + "pairwise_causal_calibration_pairs": 22.0, + "predicted_margin": -0.37508336216163984, + "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": -0.08708576405217451 + }, + "1822210471/train0": { + "base_success": 0.0, + "base_utility": 0.09442546218633652, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.6785714285714286, + "pairwise_causal_calibration_confidence": 0.5293359255305599, + "pairwise_causal_calibration_ece": 0.14923550304086874, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.11426068280426432, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.03802649304270744, + "selected_candidate_index": 5.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.09442546218633652, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9619735069572926, + "tau": -0.08708576405217451 + }, + "1822210471/train1": { + "base_success": 0.0, + "base_utility": 0.09442546218633652, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_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": 0.5263410515594875, + "pairwise_causal_calibration_ece": 0.04508751986908388, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.1660994968065328, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.07539409399032593, + "selected_candidate_index": 5.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.09442546218633652, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.924605906009674, + "tau": -0.08708576405217451 + }, + "1822210471/train2": { + "base_success": 0.0, + "base_utility": 0.09442546218633652, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "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": 0.5328646888714765, + "pairwise_causal_calibration_ece": 0.10999245398566637, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.12061726254527255, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.058819372206926346, + "selected_candidate_index": 3.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.09442546218633652, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9411806277930737, + "tau": -0.08708576405217451 + }, + "1845489013/train0": { + "base_success": 0.0, + "base_utility": 0.11255800724029541, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_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": 0.526177933898391, + "pairwise_causal_calibration_ece": 0.09760650532696241, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.18030681051640662, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.0428844690322876, + "selected_candidate_index": 0.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.11255800724029541, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9571155309677124, + "tau": -0.08708576405217451 + }, + "1845489013/train1": { + "base_success": 0.0, + "base_utility": 0.11255800724029541, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.6785714285714286, + "pairwise_causal_calibration_confidence": 0.5244266428569392, + "pairwise_causal_calibration_ece": 0.15414478571448942, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.1558320961817238, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.09529292583465576, + "selected_candidate_index": 5.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.11255800724029541, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9047070741653442, + "tau": -0.08708576405217451 + }, + "1845489013/train2": { + "base_success": 0.0, + "base_utility": 0.11255800724029541, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.75, + "pairwise_causal_calibration_confidence": 0.5261804541244063, + "pairwise_causal_calibration_ece": 0.22381954587559372, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.15437862688701337, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.0861951932311058, + "selected_candidate_index": 3.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.11255800724029541, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9138048067688942, + "tau": -0.08708576405217451 + }, + "1999022574/train0": { + "base_success": 0.0, + "base_utility": 0.8279688954353333, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.6785714285714286, + "pairwise_causal_calibration_confidence": 0.5240420866010768, + "pairwise_causal_calibration_ece": 0.15452934197035184, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.010543651527821427, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.5970457196235657, + "selected_candidate_index": 1.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.5970457196235657, + "selected_utility_gain_over_base": -0.23092317581176758, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.4029542803764343, + "tau": -0.08708576405217451 + }, + "1999022574/train1": { + "base_success": 0.0, + "base_utility": 0.8279688954353333, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_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": 0.5293282193230285, + "pairwise_causal_calibration_ece": 0.1364710764658857, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.09841891022097132, + "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": 0.8279688954353333, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.1720311045646667, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": -0.08708576405217451 + }, + "1999022574/train2": { + "base_success": 0.0, + "base_utility": 0.8279688954353333, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.6785714285714286, + "pairwise_causal_calibration_confidence": 0.5292740866826795, + "pairwise_causal_calibration_ece": 0.1492973418887491, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.11507066220959525, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.7970014810562134, + "selected_candidate_index": 2.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.8279688954353333, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.2029985189437866, + "tau": -0.08708576405217451 + }, + "2012211287/train0": { + "base_success": 0.0, + "base_utility": 0.7282466888427734, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "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": 0.5059729607314479, + "pairwise_causal_calibration_ece": 0.13688418212569498, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.32231182251355883, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.7277753949165344, + "selected_candidate_index": 2.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.7282466888427734, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.2722246050834656, + "tau": -0.08708576405217451 + }, + "2012211287/train1": { + "base_success": 0.0, + "base_utility": 0.7282466888427734, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.7037037037037037, + "pairwise_causal_calibration_confidence": 0.5090337381829941, + "pairwise_causal_calibration_ece": 0.1946699655207096, + "pairwise_causal_calibration_pairs": 27.0, + "predicted_margin": -0.3136591221973332, + "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": 0.7282466888427734, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.2717533111572266, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": -0.08708576405217451 + }, + "2012211287/train2": { + "base_success": 0.0, + "base_utility": 0.7282466888427734, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.8571428571428571, + "pairwise_causal_calibration_confidence": 0.5098518787108952, + "pairwise_causal_calibration_ece": 0.3472909784319619, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.21985746212543253, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.7270426154136658, + "selected_candidate_index": 4.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.7282466888427734, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.2729573845863342, + "tau": -0.08708576405217451 + }, + "2038594191/train0": { + "base_success": 0.0, + "base_utility": 0.09504059702157974, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.7857142857142857, + "pairwise_causal_calibration_confidence": 0.5405964043309816, + "pairwise_causal_calibration_ece": 0.24511788138330415, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.064572249902453, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.15247176587581635, + "selected_candidate_index": 7.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.15247176587581635, + "selected_utility_gain_over_base": 0.0574311688542366, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8475282341241837, + "tau": -0.08708576405217451 + }, + "2038594191/train1": { + "base_success": 0.0, + "base_utility": 0.09504059702157974, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.32142857142857145, + "pairwise_causal_calibration_confidence": 0.5321023921216932, + "pairwise_causal_calibration_ece": 0.21067382069312174, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.0309942873094304, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.1902807205915451, + "selected_candidate_index": 4.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.14092111587524414, + "selected_utility_gain_over_base": 0.0458805188536644, + "selector_regret": 0.049359604716300964, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.809719279408455, + "tau": -0.08708576405217451 + }, + "2038594191/train2": { + "base_success": 0.0, + "base_utility": 0.09504059702157974, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5357142857142857, + "pairwise_causal_calibration_confidence": 0.5393910100975016, + "pairwise_causal_calibration_ece": 0.003676724383215868, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.09610840435854003, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.24880830943584442, + "selected_candidate_index": 7.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.23134692013263702, + "selected_utility_gain_over_base": 0.13630632311105728, + "selector_regret": 0.017461389303207397, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7511916905641556, + "tau": -0.08708576405217451 + }, + "2054899126/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": 0.39285714285714285, + "pairwise_causal_calibration_confidence": 0.5044044230750341, + "pairwise_causal_calibration_ece": 0.055773640108945616, + "pairwise_causal_calibration_pairs": 14.0, + "predicted_margin": -0.43322840250453093, + "proposal_oracle_success": 0.5, + "proposal_oracle_utility": 1.0726270377635956, + "selected_candidate_index": 4.5, + "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": 0.5, + "support_gap": 0.9273729622364044, + "tau": -0.08708576405217451 + }, + "2054899126/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": 0.25, + "pairwise_causal_calibration_confidence": 0.509379035536545, + "pairwise_causal_calibration_ece": 0.12968951776827248, + "pairwise_causal_calibration_pairs": 14.0, + "predicted_margin": -0.4369772845393875, + "proposal_oracle_success": 0.5, + "proposal_oracle_utility": 1.0755530372262, + "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": 0.5, + "support_gap": 0.9244469627737999, + "tau": -0.08708576405217451 + }, + "2054899126/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": 0.5065285020074936, + "pairwise_causal_calibration_ece": 0.05030717756768177, + "pairwise_causal_calibration_pairs": 14.0, + "predicted_margin": -0.4175765945025155, + "proposal_oracle_success": 0.5, + "proposal_oracle_utility": 1.0683744624257088, + "selected_candidate_index": 0.0, + "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": 0.5, + "support_gap": 0.9316255375742912, + "tau": -0.08708576405217451 + }, + "2061635888/train0": { + "base_success": 0.0, + "base_utility": 0.08620692789554596, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.7142857142857143, + "pairwise_causal_calibration_confidence": 0.5348410330504886, + "pairwise_causal_calibration_ece": 0.17944468123522572, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.07878378828125482, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.07642464339733124, + "selected_candidate_index": 0.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.07517768442630768, + "selected_utility_gain_over_base": -0.011029243469238281, + "selector_regret": 0.0012469589710235596, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9235753566026688, + "tau": -0.08708576405217451 + }, + "2061635888/train1": { + "base_success": 0.0, + "base_utility": 0.08620692789554596, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "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": 0.5355655008647758, + "pairwise_causal_calibration_ece": 0.21443449913522417, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.004872853139381891, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.09046469628810883, + "selected_candidate_index": 4.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.08908586204051971, + "selected_utility_gain_over_base": 0.002878934144973755, + "selector_regret": 0.0013788342475891113, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9095353037118912, + "tau": -0.08708576405217451 + }, + "2061635888/train2": { + "base_success": 0.0, + "base_utility": 0.08620692789554596, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.42857142857142855, + "pairwise_causal_calibration_confidence": 0.5378360101559639, + "pairwise_causal_calibration_ece": 0.10926458158453539, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.061471356218488246, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.07902413606643677, + "selected_candidate_index": 2.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.07194386422634125, + "selected_utility_gain_over_base": -0.014263063669204712, + "selector_regret": 0.00708027184009552, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9209758639335632, + "tau": -0.08708576405217451 + }, + "211245842/train0": { + "base_success": 0.0, + "base_utility": 0.18430444598197937, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_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": 0.5137063430075536, + "pairwise_causal_calibration_ece": 0.013706343007553579, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.127498319380692, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.5822281837463379, + "selected_candidate_index": 2.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.18430444598197937, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.3979237377643585, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.417771816253662, + "tau": -0.08708576405217451 + }, + "211245842/train1": { + "base_success": 0.0, + "base_utility": 0.18430444598197937, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_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": 0.5137524311865478, + "pairwise_causal_calibration_ece": 0.12089528832940494, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.25192844397522873, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.2016763538122177, + "selected_candidate_index": 2.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.18430444598197937, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.017371907830238342, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7983236461877823, + "tau": -0.08708576405217451 + }, + "211245842/train2": { + "base_success": 0.0, + "base_utility": 0.18430444598197937, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_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": 0.5176140630711756, + "pairwise_causal_calibration_ece": 0.017614063071175634, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.19743010297696217, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.2142850160598755, + "selected_candidate_index": 4.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.18430444598197937, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.029980570077896118, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7857149839401245, + "tau": -0.08708576405217451 + }, + "278008524/train0": { + "base_success": 0.5, + "base_utility": 1.1250072121620178, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.5, + "hidden_chart_oracle_utility": 1.1139086410403252, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5714285714285714, + "pairwise_causal_calibration_confidence": 0.5171374463367512, + "pairwise_causal_calibration_ece": 0.027145562545910107, + "pairwise_causal_calibration_pairs": 14.0, + "predicted_margin": -0.2659005496346014, + "proposal_oracle_success": 0.5, + "proposal_oracle_utility": 1.0722030997276306, + "selected_candidate_index": 3.0, + "selected_success": 0.5, + "selected_success_gain_over_base": 0.0, + "selected_utility": 1.1250072121620178, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.04170554131269455, + "tau": -0.08708576405217451 + }, + "278008524/train1": { + "base_success": 0.5, + "base_utility": 1.1250072121620178, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.5, + "hidden_chart_oracle_utility": 1.1139086410403252, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.4642857142857143, + "pairwise_causal_calibration_confidence": 0.517634766627836, + "pairwise_causal_calibration_ece": 0.026674526171060864, + "pairwise_causal_calibration_pairs": 14.0, + "predicted_margin": -0.2704948697931074, + "proposal_oracle_success": 0.5, + "proposal_oracle_utility": 1.1246033310890198, + "selected_candidate_index": 2.5, + "selected_success": 0.5, + "selected_success_gain_over_base": 0.0, + "selected_utility": 1.1250072121620178, + "selected_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": -0.08708576405217451 + }, + "278008524/train2": { + "base_success": 0.5, + "base_utility": 1.1250072121620178, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.5, + "hidden_chart_oracle_utility": 1.1139086410403252, + "outcome_ptr": 0.5, + "pairwise_causal_calibration_accuracy": 0.32142857142857145, + "pairwise_causal_calibration_confidence": 0.5156802850563914, + "pairwise_causal_calibration_ece": 0.09712585681390998, + "pairwise_causal_calibration_pairs": 14.0, + "predicted_margin": -0.313945285637824, + "proposal_oracle_success": 0.5, + "proposal_oracle_utility": 1.157032236456871, + "selected_candidate_index": 2.5, + "selected_success": 0.5, + "selected_success_gain_over_base": 0.0, + "selected_utility": 1.1250072121620178, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.03202502429485321, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": -0.08708576405217451 + }, + "307068879/train0": { + "base_success": 0.0, + "base_utility": 0.20915235579013824, + "coverage": 1.0, + "execute_generated": 1.0, + "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": 0.5359843336125516, + "pairwise_causal_calibration_ece": 0.0359843336125516, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.05797252706199239, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.20280277729034424, + "selected_candidate_index": 6.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.19984664022922516, + "selected_utility_gain_over_base": -0.009305715560913086, + "selector_regret": 0.0029561370611190796, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7971972227096558, + "tau": -0.08708576405217451 + }, + "307068879/train1": { + "base_success": 0.0, + "base_utility": 0.20915235579013824, + "coverage": 1.0, + "execute_generated": 1.0, + "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": 0.5411752591475067, + "pairwise_causal_calibration_ece": 0.06596759799535039, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.018932161107400383, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.20222051441669464, + "selected_candidate_index": 1.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.20222051441669464, + "selected_utility_gain_over_base": -0.0069318413734436035, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7977794855833054, + "tau": -0.08708576405217451 + }, + "307068879/train2": { + "base_success": 0.0, + "base_utility": 0.20915235579013824, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.6785714285714286, + "pairwise_causal_calibration_confidence": 0.536058056899533, + "pairwise_causal_calibration_ece": 0.1425133716718956, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.0032027848618630603, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.2035302072763443, + "selected_candidate_index": 4.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.2035302072763443, + "selected_utility_gain_over_base": -0.005622148513793945, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7964697927236557, + "tau": -0.08708576405217451 + }, + "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": 0.35714285714285715, + "pairwise_causal_calibration_confidence": 0.5098254980752558, + "pairwise_causal_calibration_ece": 0.15268264093239864, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.14882605428987036, + "proposal_oracle_success": 1.0, + "proposal_oracle_utility": 2.0, + "selected_candidate_index": 0.0, + "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": -0.08708576405217451 + }, + "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": 0.75, + "pairwise_causal_calibration_confidence": 0.5128844404712761, + "pairwise_causal_calibration_ece": 0.23711555952872387, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.26977222222104225, + "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": -0.08708576405217451 + }, + "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": 0.25, + "pairwise_causal_calibration_confidence": 0.5173905022150012, + "pairwise_causal_calibration_ece": 0.26739050221500116, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.24386219595665565, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.8275061845779419, + "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": 1.0, + "support_gap": 1.172493815422058, + "tau": -0.08708576405217451 + }, + "440577711/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": 0.8214285714285714, + "pairwise_causal_calibration_confidence": 0.5045350428116284, + "pairwise_causal_calibration_ece": 0.316893528616943, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.41396662095926473, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.2057151049375534, + "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": 1.0, + "support_gap": 1.7942848950624466, + "tau": -0.08708576405217451 + }, + "440577711/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": 0.38461538461538464, + "pairwise_causal_calibration_confidence": 0.5045702963969247, + "pairwise_causal_calibration_ece": 0.11995491178154011, + "pairwise_causal_calibration_pairs": 13.0, + "predicted_margin": -0.549285469979439, + "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": -0.08708576405217451 + }, + "440577711/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_ece": 0.0, + "pairwise_causal_calibration_pairs": 0.0, + "predicted_margin": -0.3642232105295439, + "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": -0.08708576405217451 + }, + "441046872/train0": { + "base_success": 0.0, + "base_utility": 0.2097676396369934, + "coverage": 1.0, + "execute_generated": 1.0, + "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": 0.5358891522669097, + "pairwise_causal_calibration_ece": 0.10696799059023321, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.06101528493441861, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.203963503241539, + "selected_candidate_index": 7.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.203963503241539, + "selected_utility_gain_over_base": -0.005804136395454407, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.796036496758461, + "tau": -0.08708576405217451 + }, + "441046872/train1": { + "base_success": 0.0, + "base_utility": 0.2097676396369934, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.75, + "pairwise_causal_calibration_confidence": 0.5366361005155011, + "pairwise_causal_calibration_ece": 0.21336389948449885, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.018316727382166484, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.20609354972839355, + "selected_candidate_index": 7.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.20609354972839355, + "selected_utility_gain_over_base": -0.0036740899085998535, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7939064502716064, + "tau": -0.08708576405217451 + }, + "441046872/train2": { + "base_success": 0.0, + "base_utility": 0.2097676396369934, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.6785714285714286, + "pairwise_causal_calibration_confidence": 0.5347103383915133, + "pairwise_causal_calibration_ece": 0.14386109017991533, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.037788380273409077, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.2042645812034607, + "selected_candidate_index": 6.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.2042645812034607, + "selected_utility_gain_over_base": -0.005503058433532715, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.7957354187965393, + "tau": -0.08708576405217451 + }, + "509430392/train0": { + "base_success": 0.0, + "base_utility": 0.04073460400104523, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6, + "pairwise_causal_calibration_confidence": 0.5098428001104167, + "pairwise_causal_calibration_ece": 0.09015719988958326, + "pairwise_causal_calibration_pairs": 25.0, + "predicted_margin": -0.426096250608377, + "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": 0.04073460400104523, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.9592653959989548, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": -0.08708576405217451 + }, + "509430392/train1": { + "base_success": 0.0, + "base_utility": 0.04073460400104523, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.7272727272727273, + "pairwise_causal_calibration_confidence": 0.5095947800098043, + "pairwise_causal_calibration_ece": 0.217677947262923, + "pairwise_causal_calibration_pairs": 22.0, + "predicted_margin": -0.42675791108199884, + "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": 0.04073460400104523, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.9592653959989548, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": -0.08708576405217451 + }, + "509430392/train2": { + "base_success": 0.0, + "base_utility": 0.04073460400104523, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.6818181818181818, + "pairwise_causal_calibration_confidence": 0.5074631947694757, + "pairwise_causal_calibration_ece": 0.1743549870487061, + "pairwise_causal_calibration_pairs": 22.0, + "predicted_margin": -0.32887303074558316, + "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": 0.04073460400104523, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.9592653959989548, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": -0.08708576405217451 + }, + "585904130/train0": { + "base_success": 0.0, + "base_utility": 0.07103653997182846, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_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": 0.5199574926341867, + "pairwise_causal_calibration_ece": 0.23004250736581333, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.1685022894131509, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.16448359191417694, + "selected_candidate_index": 7.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.07103653997182846, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.09344705194234848, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.835516408085823, + "tau": -0.08708576405217451 + }, + "585904130/train1": { + "base_success": 0.0, + "base_utility": 0.07103653997182846, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_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": 0.5194057469796561, + "pairwise_causal_calibration_ece": 0.019405746979656135, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.1881172524610201, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.15698926150798798, + "selected_candidate_index": 3.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.07103653997182846, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.08595272153615952, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.843010738492012, + "tau": -0.08708576405217451 + }, + "585904130/train2": { + "base_success": 0.0, + "base_utility": 0.07103653997182846, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_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": 0.5199065927971106, + "pairwise_causal_calibration_ece": 0.09133516422568205, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.014285494888099215, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.2938849627971649, + "selected_candidate_index": 5.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.19301573932170868, + "selected_utility_gain_over_base": 0.12197919934988022, + "selector_regret": 0.10086922347545624, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.706115037202835, + "tau": -0.08708576405217451 + }, + "608016290/train0": { + "base_success": 0.0, + "base_utility": 0.525360107421875, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_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": 0.5146805911324133, + "pairwise_causal_calibration_ece": 0.12182344827527042, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.2021012748656509, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.6124615669250488, + "selected_candidate_index": 7.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.525360107421875, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.08710145950317383, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.3875384330749512, + "tau": -0.08708576405217451 + }, + "608016290/train1": { + "base_success": 0.0, + "base_utility": 0.525360107421875, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_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": 0.5137128125233418, + "pairwise_causal_calibration_ece": 0.049427098237627476, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.22391340049944683, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.6101007461547852, + "selected_candidate_index": 3.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.525360107421875, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.08474063873291016, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.3898992538452148, + "tau": -0.08708576405217451 + }, + "608016290/train2": { + "base_success": 0.0, + "base_utility": 0.525360107421875, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_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": 0.5159644162835159, + "pairwise_causal_calibration_ece": 0.09117844085934124, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.1967433702484556, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.5705108642578125, + "selected_candidate_index": 3.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.525360107421875, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0451507568359375, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.4294891357421875, + "tau": -0.08708576405217451 + }, + "626486813/train0": { + "base_success": 0.0, + "base_utility": 0.04174089431762695, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.16558238118886948, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.4464285714285714, + "pairwise_causal_calibration_confidence": 0.5125775759808189, + "pairwise_causal_calibration_ece": 0.4074440941676472, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.39959073191011746, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.05774742551147938, + "selected_candidate_index": 2.5, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.04174089431762695, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.016006531193852425, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.1078349556773901, + "tau": -0.08708576405217451 + }, + "626486813/train1": { + "base_success": 0.0, + "base_utility": 0.04174089431762695, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.16558238118886948, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5892857142857143, + "pairwise_causal_calibration_confidence": 0.5139926470970907, + "pairwise_causal_calibration_ece": 0.263654494645268, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.403655648428104, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.0619367565959692, + "selected_candidate_index": 4.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.04174089431762695, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.020195862278342247, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.10364562459290028, + "tau": -0.08708576405217451 + }, + "626486813/train2": { + "base_success": 0.0, + "base_utility": 0.04174089431762695, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.16558238118886948, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5357142857142857, + "pairwise_causal_calibration_confidence": 0.5130125362779773, + "pairwise_causal_calibration_ece": 0.18529023325047886, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.26702454174718965, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.08006523177027702, + "selected_candidate_index": 5.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.04174089431762695, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.03832433745265007, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.08551714941859245, + "tau": -0.08708576405217451 + }, + "636680137/train0": { + "base_success": 0.0, + "base_utility": 0.11957359313964844, + "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": 0.6071428571428571, + "pairwise_causal_calibration_confidence": 0.5078235229574811, + "pairwise_causal_calibration_ece": 0.09931933418537597, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.4612658703790792, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.10778365284204483, + "selected_candidate_index": 5.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.11957359313964844, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.06278067082166672, + "tau": -0.08708576405217451 + }, + "636680137/train1": { + "base_success": 0.0, + "base_utility": 0.11957359313964844, + "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": 0.2857142857142857, + "pairwise_causal_calibration_confidence": 0.5062252429740812, + "pairwise_causal_calibration_ece": 0.22051095725979553, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.4800313079785443, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.11253269761800766, + "selected_candidate_index": 3.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.11957359313964844, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.05803162604570389, + "tau": -0.08708576405217451 + }, + "636680137/train2": { + "base_success": 0.0, + "base_utility": 0.11957359313964844, + "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": 1.0, + "pairwise_causal_calibration_accuracy": 0.75, + "pairwise_causal_calibration_confidence": 0.5088107099176903, + "pairwise_causal_calibration_ece": 0.24118929008230972, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.42177891564909986, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.1247115358710289, + "selected_candidate_index": 1.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.11957359313964844, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.005137942731380463, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.04585278779268265, + "tau": -0.08708576405217451 + }, + "655884605/train0": { + "base_success": 0.0, + "base_utility": 0.1440753936767578, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_rate": 0.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.7142857142857143, + "pairwise_causal_calibration_confidence": 0.5355109972920885, + "pairwise_causal_calibration_ece": 0.1787747169936258, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.06928250847505647, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.15739890933036804, + "selected_candidate_index": 7.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.13916455209255219, + "selected_utility_gain_over_base": -0.0049108415842056274, + "selector_regret": 0.018234357237815857, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.842601090669632, + "tau": -0.08708576405217451 + }, + "655884605/train1": { + "base_success": 0.0, + "base_utility": 0.1440753936767578, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_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": 0.5393068679380685, + "pairwise_causal_calibration_ece": 0.11073543936663993, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.07124935757530594, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.15038728713989258, + "selected_candidate_index": 6.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.12485501170158386, + "selected_utility_gain_over_base": -0.01922038197517395, + "selector_regret": 0.025532275438308716, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8496127128601074, + "tau": -0.08708576405217451 + }, + "655884605/train2": { + "base_success": 0.0, + "base_utility": 0.1440753936767578, + "coverage": 1.0, + "execute_generated": 1.0, + "fallback_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": 0.5402960667817187, + "pairwise_causal_calibration_ece": 0.04029606678171871, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.0047442677921717205, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.17242512106895447, + "selected_candidate_index": 6.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.1574591100215912, + "selected_utility_gain_over_base": 0.013383716344833374, + "selector_regret": 0.014966011047363281, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.8275748789310455, + "tau": -0.08708576405217451 + }, + "672328262/train0": { + "base_success": 0.0, + "base_utility": 0.10593423992395401, + "coverage": 1.0, + "execute_generated": 1.0, + "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": 0.5399196041851851, + "pairwise_causal_calibration_ece": 0.10293753867195776, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.008075281416847935, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.05979333445429802, + "selected_candidate_index": 2.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.045009810477495193, + "selected_utility_gain_over_base": -0.060924429446458817, + "selector_regret": 0.014783523976802826, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.940206665545702, + "tau": -0.08708576405217451 + }, + "672328262/train1": { + "base_success": 0.0, + "base_utility": 0.10593423992395401, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_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": 0.5306629314646217, + "pairwise_causal_calibration_ece": 0.17352007432176458, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.17563335554996382, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.053388457745313644, + "selected_candidate_index": 6.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.10593423992395401, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9466115422546864, + "tau": -0.08708576405217451 + }, + "672328262/train2": { + "base_success": 0.0, + "base_utility": 0.10593423992395401, + "coverage": 1.0, + "execute_generated": 1.0, + "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": 0.5327362714264482, + "pairwise_causal_calibration_ece": 0.032736271426448194, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.014732418206182048, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.07216622680425644, + "selected_candidate_index": 6.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.05270739272236824, + "selected_utility_gain_over_base": -0.05322684720158577, + "selector_regret": 0.0194588340818882, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.9278337731957436, + "tau": -0.08708576405217451 + }, + "701667486/train0": { + "base_success": 0.0, + "base_utility": 0.7912740707397461, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_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": 0.5070868155940906, + "pairwise_causal_calibration_ece": 0.10005604154876646, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.23997313317310848, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.7879213094711304, + "selected_candidate_index": 1.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.7912740707397461, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.2120786905288696, + "tau": -0.08708576405217451 + }, + "701667486/train1": { + "base_success": 0.0, + "base_utility": 0.7912740707397461, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 1.0, + "pairwise_causal_calibration_accuracy": 0.5357142857142857, + "pairwise_causal_calibration_confidence": 0.5088979213620758, + "pairwise_causal_calibration_ece": 0.02681636435220991, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.2717920123841143, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.7947455048561096, + "selected_candidate_index": 2.0, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.7912740707397461, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.0034714341163635254, + "success_selector_gap": 0.0, + "success_support_gap": 1.0, + "support_gap": 1.2052544951438904, + "tau": -0.08708576405217451 + }, + "701667486/train2": { + "base_success": 0.0, + "base_utility": 0.7912740707397461, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_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": 0.5112333329328568, + "pairwise_causal_calibration_ece": 0.09590952421000032, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.34033401997115287, + "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": 0.7912740707397461, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 1.208725929260254, + "success_selector_gap": 1.0, + "success_support_gap": 0.0, + "support_gap": 0.0, + "tau": -0.08708576405217451 + }, + "747260437/train0": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2786632478237152, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.42857142857142855, + "pairwise_causal_calibration_confidence": 0.5138872754655867, + "pairwise_causal_calibration_ece": 0.08531584689415811, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.3043353554007297, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.19230616092681885, + "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": 0.08635708689689636, + "tau": -0.08708576405217451 + }, + "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": 0.8928571428571429, + "pairwise_causal_calibration_confidence": 0.5196244343388416, + "pairwise_causal_calibration_ece": 0.37323270851830126, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.41501670336545593, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.1342230886220932, + "selected_candidate_index": 0.0, + "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": 0.144440159201622, + "tau": -0.08708576405217451 + }, + "747260437/train2": { + "base_success": 1.0, + "base_utility": 2.0, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.0, + "hidden_chart_oracle_utility": 0.2786632478237152, + "outcome_ptr": 0.0, + "pairwise_causal_calibration_accuracy": 0.5357142857142857, + "pairwise_causal_calibration_confidence": 0.5073099404164519, + "pairwise_causal_calibration_ece": 0.028404345297833755, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.41403821961369863, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.18527252972126007, + "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": 0.09339071810245514, + "tau": -0.08708576405217451 + }, + "750448296/train0": { + "base_success": 0.3333333333333333, + "base_utility": 0.7605033268531164, + "coverage": 0.3333333333333333, + "execute_generated": 0.3333333333333333, + "fallback_rate": 0.6666666666666666, + "hidden_chart_oracle_success": 0.6666666666666666, + "hidden_chart_oracle_utility": 1.4079964955647786, + "outcome_ptr": 0.3333333333333333, + "pairwise_causal_calibration_accuracy": 0.42857142857142855, + "pairwise_causal_calibration_confidence": 0.5198543967759743, + "pairwise_causal_calibration_ece": 0.1606577796861187, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.18584857501276866, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.20943730572859445, + "selected_candidate_index": 5.666666666666667, + "selected_success": 0.3333333333333333, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.8105894525845846, + "selected_utility_gain_over_base": 0.0500861257314682, + "selector_regret": 0.013355518380800882, + "success_selector_gap": 0.0, + "success_support_gap": 0.6666666666666666, + "support_gap": 1.1985591898361843, + "tau": -0.08708576405217451 + }, + "750448296/train1": { + "base_success": 0.3333333333333333, + "base_utility": 0.7605033268531164, + "coverage": 0.3333333333333333, + "execute_generated": 0.3333333333333333, + "fallback_rate": 0.6666666666666666, + "hidden_chart_oracle_success": 0.6666666666666666, + "hidden_chart_oracle_utility": 1.4079964955647786, + "outcome_ptr": 0.6666666666666666, + "pairwise_causal_calibration_accuracy": 0.39285714285714285, + "pairwise_causal_calibration_confidence": 0.5196228117757805, + "pairwise_causal_calibration_ece": 0.12676566891863764, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.11894762128301756, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.22532296180725098, + "selected_candidate_index": 3.3333333333333335, + "selected_success": 0.3333333333333333, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.7964938481648763, + "selected_utility_gain_over_base": 0.03599052131175995, + "selector_regret": 0.027911449472109478, + "success_selector_gap": 0.0, + "success_support_gap": 0.6666666666666666, + "support_gap": 1.1875135501225789, + "tau": -0.08708576405217451 + }, + "750448296/train2": { + "base_success": 0.3333333333333333, + "base_utility": 0.7605033268531164, + "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": 0.3333333333333333, + "pairwise_causal_calibration_accuracy": 0.369047619047619, + "pairwise_causal_calibration_confidence": 0.5195045055291346, + "pairwise_causal_calibration_ece": 0.15045688648151567, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.1279298521209656, + "proposal_oracle_success": 0.0, + "proposal_oracle_utility": 0.2481476068496704, + "selected_candidate_index": 3.0, + "selected_success": 0.0, + "selected_success_gain_over_base": -0.3333333333333333, + "selected_utility": 0.1862319732705752, + "selected_utility_gain_over_base": -0.5742713535825411, + "selector_regret": 0.07596839219331741, + "success_selector_gap": 0.0, + "success_support_gap": 0.6666666666666666, + "support_gap": 1.1598488887151082, + "tau": -0.08708576405217451 + }, + "767747424/train0": { + "base_success": 0.5, + "base_utility": 1.0747087560594082, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.25, + "hidden_chart_oracle_utility": 0.6737305596470833, + "outcome_ptr": 0.25, + "pairwise_causal_calibration_accuracy": 0.4017857142857143, + "pairwise_causal_calibration_confidence": 0.5118569560047292, + "pairwise_causal_calibration_ece": 0.1902881683926222, + "pairwise_causal_calibration_pairs": 22.75, + "predicted_margin": -0.3105746772245132, + "proposal_oracle_success": 0.25, + "proposal_oracle_utility": 0.5789013803005219, + "selected_candidate_index": 2.25, + "selected_success": 0.5, + "selected_success_gain_over_base": 0.0, + "selected_utility": 1.0747087560594082, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.019455675035715103, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.10475751012563705, + "tau": -0.08708576405217451 + }, + "767747424/train1": { + "base_success": 0.5, + "base_utility": 1.0747087560594082, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.25, + "hidden_chart_oracle_utility": 0.6737305596470833, + "outcome_ptr": 0.25, + "pairwise_causal_calibration_accuracy": 0.4553571428571429, + "pairwise_causal_calibration_confidence": 0.5130660545319694, + "pairwise_causal_calibration_ece": 0.23038223964120388, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.3587369662558639, + "proposal_oracle_success": 0.25, + "proposal_oracle_utility": 0.5825987188145518, + "selected_candidate_index": 2.75, + "selected_success": 0.5, + "selected_success_gain_over_base": 0.0, + "selected_utility": 1.0747087560594082, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.019207116216421127, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.1008116127923131, + "tau": -0.08708576405217451 + }, + "767747424/train2": { + "base_success": 0.5, + "base_utility": 1.0747087560594082, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.25, + "hidden_chart_oracle_utility": 0.6737305596470833, + "outcome_ptr": 0.25, + "pairwise_causal_calibration_accuracy": 0.4910714285714286, + "pairwise_causal_calibration_confidence": 0.5118319733441975, + "pairwise_causal_calibration_ece": 0.205693305843048, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": -0.35684179754966683, + "proposal_oracle_success": 0.25, + "proposal_oracle_utility": 0.5809952197596431, + "selected_candidate_index": 3.5, + "selected_success": 0.5, + "selected_success_gain_over_base": 0.0, + "selected_utility": 1.0747087560594082, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.013884376734495163, + "success_selector_gap": 0.0, + "success_support_gap": 0.0, + "support_gap": 0.09709237236529589, + "tau": -0.08708576405217451 + } +} diff --git a/workspace/runs/ctt_base_context_obs_learned_dominance_basic_perdim_trainmax_train_to_test/metrics_by_task.json b/workspace/runs/ctt_base_context_obs_learned_dominance_basic_perdim_trainmax_train_to_test/metrics_by_task.json new file mode 100644 index 0000000000000000000000000000000000000000..c8c918ee6a9534c3a1126c1a07409e0016087d81 --- /dev/null +++ b/workspace/runs/ctt_base_context_obs_learned_dominance_basic_perdim_trainmax_train_to_test/metrics_by_task.json @@ -0,0 +1,137 @@ +{ + "LiftPegUpright-v1": { + "base_success": 0.0, + "base_utility": 0.4796089559793472, + "coverage": 0.4, + "execute_generated": 0.4, + "fallback_rate": 0.6, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.3333333333333333, + "pairwise_causal_calibration_accuracy": 0.6016754850088184, + "pairwise_causal_calibration_confidence": 0.5229362359705314, + "pairwise_causal_calibration_ece": 0.11584970123844203, + "pairwise_causal_calibration_pairs": 27.933333333333334, + "predicted_margin": -0.14503053505926972, + "proposal_oracle_success": 0.13333333333333333, + "proposal_oracle_utility": 0.6601470321416855, + "selected_candidate_index": 4.333333333333333, + "selected_success": 0.0, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.4771528899669647, + "selected_utility_gain_over_base": -0.002456066012382507, + "selector_regret": 0.18441228767236073, + "success_selector_gap": 0.13333333333333333, + "success_support_gap": 0.8666666666666667, + "support_gap": 1.3398529678583144, + "tau": -0.08708576405217451 + }, + "PickCube-v1": { + "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": 0.32142857142857145, + "pairwise_causal_calibration_confidence": 0.5472287677691599, + "pairwise_causal_calibration_ece": 0.22580019634058843, + "pairwise_causal_calibration_pairs": 28.0, + "predicted_margin": 0.28070211086543817, + "proposal_oracle_success": 0.3333333333333333, + "proposal_oracle_utility": 1.021850009759267, + "selected_candidate_index": 1.0, + "selected_success": 0.0, + "selected_success_gain_over_base": -1.0, + "selected_utility": 0.26197221378485364, + "selected_utility_gain_over_base": -1.7380277862151463, + "selector_regret": 0.7598777959744135, + "success_selector_gap": 0.3333333333333333, + "success_support_gap": 0.6666666666666666, + "support_gap": 0.9781499902407328, + "tau": -0.08708576405217451 + }, + "PullCube-v1": { + "base_success": 0.36, + "base_utility": 0.8032506120204925, + "coverage": 0.13333333333333333, + "execute_generated": 0.13333333333333333, + "fallback_rate": 0.8666666666666667, + "hidden_chart_oracle_success": 0.52, + "hidden_chart_oracle_utility": 1.1467878860235214, + "outcome_ptr": 0.32, + "pairwise_causal_calibration_accuracy": 0.5106143856143858, + "pairwise_causal_calibration_confidence": 0.5170252389197862, + "pairwise_causal_calibration_ece": 0.149483534146025, + "pairwise_causal_calibration_pairs": 24.16, + "predicted_margin": -0.27933737647327517, + "proposal_oracle_success": 0.17333333333333334, + "proposal_oracle_utility": 0.46170220340291657, + "selected_candidate_index": 3.2666666666666666, + "selected_success": 0.3466666666666667, + "selected_success_gain_over_base": -0.013333333333333334, + "selected_utility": 0.7870221028725306, + "selected_utility_gain_over_base": -0.016228509147961933, + "selector_regret": 0.01846348727742831, + "success_selector_gap": 0.0, + "success_support_gap": 0.3466666666666667, + "support_gap": 0.6881442584097386, + "tau": -0.08708576405217451 + }, + "PushCube-v1": { + "base_success": 0.8333333333333334, + "base_utility": 1.6734557673335075, + "coverage": 0.0, + "execute_generated": 0.0, + "fallback_rate": 1.0, + "hidden_chart_oracle_success": 0.8333333333333334, + "hidden_chart_oracle_utility": 1.7015800600250561, + "outcome_ptr": 0.16666666666666666, + "pairwise_causal_calibration_accuracy": 0.5693706076059017, + "pairwise_causal_calibration_confidence": 0.5109774849646357, + "pairwise_causal_calibration_ece": 0.1513149797579122, + "pairwise_causal_calibration_pairs": 23.833333333333332, + "predicted_margin": -0.3951348987851248, + "proposal_oracle_success": 0.5555555555555556, + "proposal_oracle_utility": 1.1790093589160178, + "selected_candidate_index": 2.5555555555555554, + "selected_success": 0.8333333333333334, + "selected_success_gain_over_base": 0.0, + "selected_utility": 1.6734557673335075, + "selected_utility_gain_over_base": 0.0, + "selector_regret": 0.32654423266649246, + "success_selector_gap": 0.16666666666666666, + "success_support_gap": 0.3333333333333333, + "support_gap": 0.6220440144340197, + "tau": -0.08708576405217451 + }, + "StackCube-v1": { + "base_success": 0.09090909090909091, + "base_utility": 0.5003613992170854, + "coverage": 0.45454545454545453, + "execute_generated": 0.45454545454545453, + "fallback_rate": 0.5454545454545454, + "hidden_chart_oracle_success": 1.0, + "hidden_chart_oracle_utility": 2.0, + "outcome_ptr": 0.6060606060606061, + "pairwise_causal_calibration_accuracy": 0.5182683982683982, + "pairwise_causal_calibration_confidence": 0.527034482028972, + "pairwise_causal_calibration_ece": 0.12741592448328593, + "pairwise_causal_calibration_pairs": 27.818181818181817, + "predicted_margin": -0.07917090287344432, + "proposal_oracle_success": 0.18181818181818182, + "proposal_oracle_utility": 0.634040683056369, + "selected_candidate_index": 2.757575757575758, + "selected_success": 0.09090909090909091, + "selected_success_gain_over_base": 0.0, + "selected_utility": 0.48857860628402594, + "selected_utility_gain_over_base": -0.011782792933059462, + "selector_regret": 0.18203219771385193, + "success_selector_gap": 0.12121212121212122, + "success_support_gap": 0.8181818181818182, + "support_gap": 1.365959316943631, + "tau": -0.08708576405217451 + } +} diff --git a/workspace/runs/ctt_base_context_obs_learned_dominance_basic_perdim_trainmax_train_to_test/split_hash.txt b/workspace/runs/ctt_base_context_obs_learned_dominance_basic_perdim_trainmax_train_to_test/split_hash.txt new file mode 100644 index 0000000000000000000000000000000000000000..865ad41e869c1bfbc57e4db65ed79e91e6d93081 --- /dev/null +++ b/workspace/runs/ctt_base_context_obs_learned_dominance_basic_perdim_trainmax_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_basic_perdim_trainmax_train_to_test/table.tex b/workspace/runs/ctt_base_context_obs_learned_dominance_basic_perdim_trainmax_train_to_test/table.tex new file mode 100644 index 0000000000000000000000000000000000000000..1990b5bc592e63f1b92739e7cb3ca3a0d98fd3fa --- /dev/null +++ b/workspace/runs/ctt_base_context_obs_learned_dominance_basic_perdim_trainmax_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 & 0.2361 & 0.7639 & 0.3333 & 0.3056 & 0.2222 & 0.3611 & 0.5139 & 0.0694 & 0.0103 \\ +\bottomrule +\end{tabular} diff --git a/workspace/runs/ctt_base_context_obs_learned_dominance_basic_perdim_trainmax_train_to_test/train.log b/workspace/runs/ctt_base_context_obs_learned_dominance_basic_perdim_trainmax_train_to_test/train.log new file mode 100644 index 0000000000000000000000000000000000000000..9c6d7ccbd5a0ec682cd6c398ac6bd46313871707 --- /dev/null +++ b/workspace/runs/ctt_base_context_obs_learned_dominance_basic_perdim_trainmax_train_to_test/train.log @@ -0,0 +1,5 @@ +trained ridge dominance calibrator on calibration measured rows only +fit_objective=pointwise +selected_lambda=0.0 +tau=-0.087086 +calibration_input=runs/ctt_base_context_obs_train_cal_perdim_trainmax_rollout_comparison/combined_measured_candidates.json