From e3011f144a018f6415b3c2a636c0646e412b3ced Mon Sep 17 00:00:00 2001 From: YurenHao0426 Date: Mon, 10 Aug 2026 10:18:26 -0500 Subject: results: select BabyAI clean configuration --- results/babyai_shared/b0_selector.json | 105 +++++++++++++++++++++++++++++++++ 1 file changed, 105 insertions(+) create mode 100644 results/babyai_shared/b0_selector.json (limited to 'results/babyai_shared/b0_selector.json') diff --git a/results/babyai_shared/b0_selector.json b/results/babyai_shared/b0_selector.json new file mode 100644 index 0000000..4368e44 --- /dev/null +++ b/results/babyai_shared/b0_selector.json @@ -0,0 +1,105 @@ +{ + "candidates": [ + { + "bp_action_accuracy": 0.9974355321270836, + "bp_mission_lesion_drop": 0.274, + "bp_mission_lesion_success": 0.704, + "bp_rollout_success": 0.978, + "checks": { + "both_finite": true, + "bp_mission_lesion_drop_at_least_0p2": true, + "bp_success_at_least_0p8": true, + "clean_kp_success_at_least_0p8": true + }, + "clean_kp_action_accuracy": 0.9974355321270836, + "clean_kp_rollout_success": 0.978, + "eligible": true, + "hidden_layers": 2, + "learning_rate": 0.01, + "source_files": [ + "results/babyai_shared/b0/d2_lr0.01_bp.json", + "results/babyai_shared/b0/d2_lr0.01_clean_kp.json" + ] + }, + { + "bp_action_accuracy": 0.9974355321270836, + "bp_mission_lesion_drop": 0.07199999999999995, + "bp_mission_lesion_success": 0.906, + "bp_rollout_success": 0.978, + "checks": { + "both_finite": true, + "bp_mission_lesion_drop_at_least_0p2": false, + "bp_success_at_least_0p8": true, + "clean_kp_success_at_least_0p8": true + }, + "clean_kp_action_accuracy": 0.9974355321270836, + "clean_kp_rollout_success": 0.978, + "eligible": false, + "hidden_layers": 2, + "learning_rate": 0.03, + "source_files": [ + "results/babyai_shared/b0/d2_lr0.03_bp.json", + "results/babyai_shared/b0/d2_lr0.03_clean_kp.json" + ] + }, + { + "bp_action_accuracy": 0.9974355321270836, + "bp_mission_lesion_drop": 0.42799999999999994, + "bp_mission_lesion_success": 0.55, + "bp_rollout_success": 0.978, + "checks": { + "both_finite": true, + "bp_mission_lesion_drop_at_least_0p2": true, + "bp_success_at_least_0p8": true, + "clean_kp_success_at_least_0p8": true + }, + "clean_kp_action_accuracy": 0.9974355321270836, + "clean_kp_rollout_success": 0.978, + "eligible": true, + "hidden_layers": 4, + "learning_rate": 0.01, + "source_files": [ + "results/babyai_shared/b0/d4_lr0.01_bp.json", + "results/babyai_shared/b0/d4_lr0.01_clean_kp.json" + ] + }, + { + "bp_action_accuracy": 0.9974355321270836, + "bp_mission_lesion_drop": 0.0, + "bp_mission_lesion_success": 0.978, + "bp_rollout_success": 0.978, + "checks": { + "both_finite": true, + "bp_mission_lesion_drop_at_least_0p2": false, + "bp_success_at_least_0p8": true, + "clean_kp_success_at_least_0p8": true + }, + "clean_kp_action_accuracy": 0.9974355321270836, + "clean_kp_rollout_success": 0.978, + "eligible": false, + "hidden_layers": 4, + "learning_rate": 0.03, + "source_files": [ + "results/babyai_shared/b0/d4_lr0.03_bp.json", + "results/babyai_shared/b0/d4_lr0.03_clean_kp.json" + ] + } + ], + "gate": "pass", + "raw_or_sdil_results_read": false, + "selected": { + "b1_epochs": 40, + "b1_model_and_shuffle_seeds": [ + 4101, + 4102, + 4103 + ], + "context_gain": 1.0, + "hidden_layers": 2, + "learning_rate": 0.01, + "width": 256 + }, + "selection_rule": "highest clean-KP rollout success, then action accuracy, then fewer layers, then smaller learning rate, among eligible rows", + "stage": "babyai_shared_b0_selector", + "test_split_generated_or_read": false +} -- cgit v1.2.3