diff options
Diffstat (limited to 'results/bci_v2_recovery_dev/bci_v2_recovery_t24_m0.json')
| -rw-r--r-- | results/bci_v2_recovery_dev/bci_v2_recovery_t24_m0.json | 430 |
1 files changed, 430 insertions, 0 deletions
diff --git a/results/bci_v2_recovery_dev/bci_v2_recovery_t24_m0.json b/results/bci_v2_recovery_dev/bci_v2_recovery_t24_m0.json new file mode 100644 index 0000000..9909867 --- /dev/null +++ b/results/bci_v2_recovery_dev/bci_v2_recovery_t24_m0.json @@ -0,0 +1,430 @@ +{ + "args": { + "model_seed": 0, + "outdir": "results/bci_v2_recovery_dev", + "task_seed": 24 + }, + "assays": { + "challenge": { + "active_state_episode_steps_by_mode": { + "acute_critic_lesion": 12158, + "acute_outcome_lesion": 12158, + "intact": 12158 + }, + "episodes_per_target": 128, + "maximum_steps_per_episode": 28, + "selection_over_targets": false, + "targets": [ + 1.55, + 1.6, + 1.65, + 1.7, + 1.75, + 1.8 + ], + "trajectory_seeds": { + "1.55": 494000, + "1.6": 494001, + "1.65": 494002, + "1.7": 494003, + "1.75": 494004, + "1.8": 494005 + } + }, + "performance_evaluation_episodes": 256, + "performance_evaluation_seed": 460024 + }, + "conditions": { + "critic_training_lesion": { + "cost": { + "active_state_episode_steps": 16221, + "cursor_scalar_observations": 8576, + "maximum_state_episode_steps": 25088, + "reverse_mode_calls": 0, + "role_probe_examples": 4288, + "task_loss_queries": 0, + "terminal_outcome_observations": 896 + }, + "critic_l2_after_training": 0.0, + "daily_success": [ + 0.0, + 0.0, + 0.0, + 0.015625, + 0.0, + 0.0, + 0.046875, + 0.140625, + 0.71875, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "early_success": 0.0, + "evaluation_active_state_episode_steps": 256, + "final_success": 1.0, + "late_success": 1.0, + "learning_gain": 1.0, + "role_cosine_after_training": 0.9901587963104248, + "training_wall_s": 0.19473905116319656 + }, + "fixed_role": { + "cost": { + "active_state_episode_steps": 25088, + "cursor_scalar_observations": 12544, + "maximum_state_episode_steps": 25088, + "reverse_mode_calls": 0, + "role_probe_examples": 6272, + "task_loss_queries": 0, + "terminal_outcome_observations": 896 + }, + "critic_l2_after_training": 0.06098082661628723, + "daily_success": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ], + "early_success": 0.0, + "evaluation_active_state_episode_steps": 7168, + "final_success": 0.0, + "late_success": 0.0, + "learning_gain": 0.0, + "role_cosine_after_training": -0.1393236517906189, + "training_wall_s": 0.23979467898607254 + }, + "intact": { + "cost": { + "active_state_episode_steps": 14085, + "cursor_scalar_observations": 7610, + "maximum_state_episode_steps": 25088, + "reverse_mode_calls": 0, + "role_probe_examples": 3805, + "task_loss_queries": 0, + "terminal_outcome_observations": 896 + }, + "critic_l2_after_training": 0.4679003357887268, + "daily_success": [ + 0.0, + 0.0, + 0.0, + 0.015625, + 0.0, + 0.03125, + 0.28125, + 0.875, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "early_success": 0.0, + "evaluation_active_state_episode_steps": 257, + "final_success": 1.0, + "late_success": 1.0, + "learning_gain": 1.0, + "role_cosine_after_training": 0.9919018745422363, + "training_wall_s": 0.27622058615088463 + }, + "oracle_role": { + "cost": { + "active_state_episode_steps": 13623, + "cursor_scalar_observations": 7408, + "maximum_state_episode_steps": 25088, + "reverse_mode_calls": 0, + "role_probe_examples": 3704, + "task_loss_queries": 0, + "terminal_outcome_observations": 896 + }, + "critic_l2_after_training": 0.47546476125717163, + "daily_success": [ + 0.0, + 0.0, + 0.0, + 0.015625, + 0.0, + 0.046875, + 0.375, + 0.984375, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0, + 1.0 + ], + "early_success": 0.0, + "evaluation_active_state_episode_steps": 257, + "final_success": 1.0, + "late_success": 1.0, + "learning_gain": 1.0, + "role_cosine_after_training": 1.0000001192092896, + "training_wall_s": 0.19996124133467674 + }, + "outcome_training_lesion": { + "cost": { + "active_state_episode_steps": 17888, + "cursor_scalar_observations": 9260, + "maximum_state_episode_steps": 25088, + "reverse_mode_calls": 0, + "role_probe_examples": 4630, + "task_loss_queries": 0, + "terminal_outcome_observations": 896 + }, + "critic_l2_after_training": 0.2022681087255478, + "daily_success": [ + 0.0, + 0.0, + 0.0, + 0.015625, + 0.0, + 0.015625, + 0.078125, + 0.234375, + 0.59375, + 0.78125, + 0.984375, + 1.0, + 1.0, + 1.0 + ], + "early_success": 0.0, + "evaluation_active_state_episode_steps": 768, + "final_success": 1.0, + "late_success": 1.0, + "learning_gain": 1.0, + "role_cosine_after_training": 0.9417572021484375, + "training_wall_s": 0.23217474296689034 + }, + "plasticity_lesion": { + "cost": { + "active_state_episode_steps": 25088, + "cursor_scalar_observations": 12544, + "maximum_state_episode_steps": 25088, + "reverse_mode_calls": 0, + "role_probe_examples": 6272, + "task_loss_queries": 0, + "terminal_outcome_observations": 896 + }, + "critic_l2_after_training": 0.062091100960969925, + "daily_success": [ + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 + ], + "early_success": 0.0, + "evaluation_active_state_episode_steps": 7168, + "final_success": 0.0, + "late_success": 0.0, + "learning_gain": 0.0, + "role_cosine_after_training": 0.9967973232269287, + "training_wall_s": 0.21114370226860046 + } + }, + "config": { + "context_ar": 0.8, + "context_dim": 16, + "coupling_scale": 1.0, + "critic_eta": 0.03, + "days": 14, + "eligibility_decay": 0.8, + "episodes_per_day": 64, + "feedback": "performance_velocity", + "forward_eta": 0.1, + "gamma": 0.8, + "inertia": 0.65, + "kappa": 0.0, + "n_background": 30, + "n_minus": 5, + "n_plus": 5, + "perturb_every": 4, + "perturb_sigma": 0.03, + "predictor_eta": 0.2, + "process_noise": 0.12, + "steps_per_episode": 28, + "target": 0.8, + "terminal_reward": 1.0, + "vectorizer_eta": 0.03, + "velocity_reward_scale": 1.0 + }, + "finite": true, + "hardware": { + "device": "cpu", + "platform": "Linux-5.15.0-161-generic-x86_64-with-glibc2.35", + "threads": 1, + "torch_version": "2.10.0+cu128" + }, + "peak_rss_mib": 996.0546875, + "protocol": { + "confirmation_seeds_touched": false, + "d4_gate_sha256": "636c587e47287338cdca7d9558dc3bfb99a2cec615badb23337025d85b8128ef", + "failed_v2_gate_sha256": "e53f46cf456d60ce4d33586ace4f4619fb1a31a58082b2b72943d4ed10cd25fd", + "fixed_config": { + "critic_eta": 0.03, + "forward_eta": 0.1, + "gamma": 0.8, + "velocity_reward_scale": 1.0 + }, + "hyperparameter_selection": false, + "name": "oral_b_v2_cold_start_recovery_development_v1", + "old_r2_gate_sha256": "4f6f969ceae88afa2523e3472a3373522991f5ecaab69440830c07479d2d3597", + "protocol_sha256": "0621b47367074cb99ee74b332192878e76acbca11678b5238e1312adc2a2bcd9", + "selection_split": "development_validation" + }, + "provenance": { + "git_commit": "dae172b1425c646f0441609c28fc5156b50f820d", + "git_tracked_dirty": false, + "input_sha256": { + "base_dynamics": "d5a373314562af0daedf55baa04b975fe643236c7821be1560d2cfb4359688fc", + "confirmation_analyzer": "29a168143266bc7078e5229183bb7ea45283973592fecd1fcd715488f5d6242b", + "confirmation_runner": "0582ae7a0c173d2fef842ea57d75b77b23cb6f70391f7013ad27ec29e0fedfdf", + "d4_gate": "636c587e47287338cdca7d9558dc3bfb99a2cec615badb23337025d85b8128ef", + "development_analyzer": "2768b877132ece6d612d2a5f1e483c495b277e26fdf54fb8a716aa8de6fd67e2", + "failed_v2_gate": "e53f46cf456d60ce4d33586ace4f4619fb1a31a58082b2b72943d4ed10cd25fd", + "old_r2_gate": "4f6f969ceae88afa2523e3472a3373522991f5ecaab69440830c07479d2d3597", + "protocol": "0621b47367074cb99ee74b332192878e76acbca11678b5238e1312adc2a2bcd9", + "recovery_metrics": "371cb6a7e2192d25ddbfad69c0ce7a1150aa1c21ca98e141bedb2afc2acbdbaa", + "runner": "cea0ef658c698c376b1fda0cf2f7f910ee44f7bcde153f2fbaf18e66a2d3661a", + "v2_dynamics": "786c5aed3091644272ae0a740913ac3cc38cdb02a8cacb6dfc8496e36836cdae", + "v2_metrics": "421da9287b92e62c742f9eeefbfdccba582c6a4ad5cf92ade4709689120fa1ee" + }, + "tracked_inputs": { + "base_dynamics": true, + "confirmation_analyzer": true, + "confirmation_runner": true, + "d4_gate": true, + "development_analyzer": true, + "failed_v2_gate": true, + "old_r2_gate": true, + "protocol": true, + "recovery_metrics": true, + "runner": true, + "v2_dynamics": true, + "v2_metrics": true + } + }, + "schema_version": 3, + "signatures": { + "acute_outcome_lesion_outcome_balanced_acc": 0.5541079585665574, + "acute_outcome_lesion_role_aligned_separation": 0.01978198133532344, + "causal_role_sign_inversion_index": 0.040138957469204, + "challenge_episodes": 768, + "challenge_failure_count": 297, + "challenge_success_count": 471, + "challenge_success_fraction": 0.61328125, + "critic_contribution_value_prediction_corr": 0.9999999999991358, + "decoder_distance_residual_corr": 0.09986939764391377, + "mean_abs_raw_soma_corr": 0.9991346687315854, + "mean_abs_residual_soma_corr": 0.05678144772056838, + "mean_critic_expectedness_contribution": 0.322402930318503, + "nonterminal_training_events": 13189, + "raw_minus_residual_abs_soma_corr": 0.942353221011017, + "role_aligned_error_cv_corr": 0.352382055212796, + "role_aligned_velocity_cv_corr": 0.9984807554798858, + "surrounding_event_decoder_balanced_acc": 0.5451291756686398, + "target_success_fraction": { + "1.55": 1.0, + "1.6": 0.9921875, + "1.65": 0.9296875, + "1.7": 0.5625, + "1.75": 0.1875, + "1.8": 0.0078125 + }, + "terminal_outcome_separation_drop_under_acute_lesion": 0.40220903209568243, + "terminal_previous_soma_outcome_balanced_acc": 0.5670719938235862, + "terminal_residual_minus_previous_soma_acc": 0.4329280061764138, + "terminal_residual_outcome_balanced_acc": 1.0, + "terminal_role_aligned_outcome_separation": 0.42199101343100587, + "terminal_training_events": 896, + "velocity_minus_error_abs_cv_corr": 0.6460987002670898 + }, + "split": "development", + "wall_s": 2.6129846908152103, + "warmup": { + "critic_training_lesion": { + "batches": 100, + "examples": 6400, + "instruction_present": false, + "predictor_max_abs_error": 2.384185791015625e-07, + "rng_seed": 602400, + "role_cosine_after_warmup": 0.9931948781013489, + "role_cursor_scalar_observations": 12800, + "role_update_enabled": true + }, + "fixed_role": { + "batches": 100, + "examples": 6400, + "instruction_present": false, + "predictor_max_abs_error": 2.384185791015625e-07, + "rng_seed": 602400, + "role_cosine_after_warmup": -0.1393236517906189, + "role_cursor_scalar_observations": 12800, + "role_update_enabled": false + }, + "intact": { + "batches": 100, + "examples": 6400, + "instruction_present": false, + "predictor_max_abs_error": 2.384185791015625e-07, + "rng_seed": 602400, + "role_cosine_after_warmup": 0.9931948781013489, + "role_cursor_scalar_observations": 12800, + "role_update_enabled": true + }, + "oracle_role": { + "batches": 100, + "examples": 6400, + "instruction_present": false, + "predictor_max_abs_error": 2.384185791015625e-07, + "rng_seed": 602400, + "role_cosine_after_warmup": 1.0000001192092896, + "role_cursor_scalar_observations": 12800, + "role_update_enabled": false + }, + "outcome_training_lesion": { + "batches": 100, + "examples": 6400, + "instruction_present": false, + "predictor_max_abs_error": 2.384185791015625e-07, + "rng_seed": 602400, + "role_cosine_after_warmup": 0.9931948781013489, + "role_cursor_scalar_observations": 12800, + "role_update_enabled": true + }, + "plasticity_lesion": { + "batches": 100, + "examples": 6400, + "instruction_present": false, + "predictor_max_abs_error": 2.384185791015625e-07, + "rng_seed": 602400, + "role_cosine_after_warmup": 0.9931948781013489, + "role_cursor_scalar_observations": 12800, + "role_update_enabled": true + } + } +} |