Auto-sync: 2026-06-29 19:11:23
Browse files- results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p10_noopbonus0p03_summary.json +412 -0
- results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p10_noopbonus0p03_summary.md +19 -0
- results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_oraclek8_summary.json +697 -0
- results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_oraclek8_summary.md +24 -0
results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p10_noopbonus0p03_summary.json
ADDED
|
@@ -0,0 +1,412 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"run_root": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs",
|
| 3 |
+
"objective": "near_miss_policy_bc5",
|
| 4 |
+
"out_name": "policy_rollout_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p10_noopbonus0p03.json",
|
| 5 |
+
"num_completed": 3,
|
| 6 |
+
"baseline_h4_policy_success": 0.2967,
|
| 7 |
+
"baseline_h16_rank_checkpoint_success": 0.29739130434782607,
|
| 8 |
+
"mean_success": 0.3466666666666667,
|
| 9 |
+
"std_success": 0.00957838356049757,
|
| 10 |
+
"mean_progress": 0.5644778893518168,
|
| 11 |
+
"mean_action_mse_to_best": 0.41778298790499135,
|
| 12 |
+
"gain_vs_h4": 0.04996666666666666,
|
| 13 |
+
"gain_vs_h16_rank_checkpoint": 0.04927536231884061,
|
| 14 |
+
"rows": [
|
| 15 |
+
{
|
| 16 |
+
"seed": 0,
|
| 17 |
+
"path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/near_miss_policy_bc5/seed_0/policy_rollout_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p10_noopbonus0p03.json",
|
| 18 |
+
"num_groups": 575,
|
| 19 |
+
"selection_mode": "retrieval_residual",
|
| 20 |
+
"num_candidates": 48,
|
| 21 |
+
"candidate_sigma": 0.0,
|
| 22 |
+
"selection_margin": 0.1,
|
| 23 |
+
"prepend_policy_candidate": false,
|
| 24 |
+
"field_optim_steps": 0,
|
| 25 |
+
"field_optim_step_size": 0.0,
|
| 26 |
+
"field_optim_trust_radius": 0.0,
|
| 27 |
+
"field_optim_l2_penalty": 0.0,
|
| 28 |
+
"retrieval_neighbors": 4,
|
| 29 |
+
"retrieval_metric": "raw",
|
| 30 |
+
"retrieval_type_min_success": 0.0,
|
| 31 |
+
"retrieval_type_success_bonus_scale": 0.0,
|
| 32 |
+
"retrieval_residual_consensus_penalty_scale": 0.0,
|
| 33 |
+
"retrieval_residual_min_source_progress": 0.0,
|
| 34 |
+
"retrieval_residual_min_source_advantage": -1000000000.0,
|
| 35 |
+
"retrieval_residual_source_progress_bonus_scale": 0.0,
|
| 36 |
+
"retrieval_residual_source_score_bonus_scale": 0.0,
|
| 37 |
+
"retrieval_residual_source_advantage_bonus_scale": 0.0,
|
| 38 |
+
"retrieval_residual_composite_l2_penalty_scale": 0.0,
|
| 39 |
+
"retrieval_residual_scale": 1.0,
|
| 40 |
+
"retrieval_residual_scales": [
|
| 41 |
+
0.35,
|
| 42 |
+
0.4,
|
| 43 |
+
0.45
|
| 44 |
+
],
|
| 45 |
+
"retrieval_residual_anchor": "expert",
|
| 46 |
+
"retrieval_residual_direction": "candidate_minus_anchor",
|
| 47 |
+
"retrieval_residual_reduce": "compose_mean_by_type",
|
| 48 |
+
"lattice_exclude_types": [
|
| 49 |
+
"residual_random_negative",
|
| 50 |
+
"residual_wrong_direction",
|
| 51 |
+
"residual_near_miss+residual_no_op"
|
| 52 |
+
],
|
| 53 |
+
"candidate_type_bonuses": {
|
| 54 |
+
"residual_no_op": 0.03
|
| 55 |
+
},
|
| 56 |
+
"candidate_type_bonus_components": false,
|
| 57 |
+
"candidate_oracle_rollouts": 0,
|
| 58 |
+
"candidate_oracle_success_rate": null,
|
| 59 |
+
"candidate_oracle_progress": null,
|
| 60 |
+
"candidate_oracle_score": null,
|
| 61 |
+
"candidate_oracle_regret": null,
|
| 62 |
+
"candidate_oracle_score_gain_over_selected": null,
|
| 63 |
+
"candidate_oracle_improvement_rate": null,
|
| 64 |
+
"candidate_oracle_selected_branch_success_rate": null,
|
| 65 |
+
"candidate_oracle_selected_branch_progress": null,
|
| 66 |
+
"candidate_oracle_type_counts": {},
|
| 67 |
+
"selected_residual_scale_counts": {
|
| 68 |
+
"0.35": 499,
|
| 69 |
+
"0.45": 69,
|
| 70 |
+
"0.4": 7
|
| 71 |
+
},
|
| 72 |
+
"policy_rollout_success_rate": 0.3356521739130435,
|
| 73 |
+
"policy_rollout_progress": 0.549884562070441,
|
| 74 |
+
"oracle_success_rate": 0.8573913043478261,
|
| 75 |
+
"action_mse_to_best": 0.40276106816433044,
|
| 76 |
+
"best_policy_val": {
|
| 77 |
+
"bc_loss": 0.13721593966086706,
|
| 78 |
+
"field_effect_loss": 0.009290305380192068,
|
| 79 |
+
"field_potential_loss": 0.2666468388504452,
|
| 80 |
+
"field_preference_loss": 0.5130573478009965,
|
| 81 |
+
"lattice_edges": 3833.3333333333335,
|
| 82 |
+
"progress_mae": 0.1933159919248687,
|
| 83 |
+
"rank_acc": 0.8265031774838766,
|
| 84 |
+
"rank_loss": 0.5130523675017886,
|
| 85 |
+
"regret_mae": 0.3756548762321472,
|
| 86 |
+
"success_accuracy": 0.8773836526605818,
|
| 87 |
+
"total_loss": 1.5581054819954767
|
| 88 |
+
},
|
| 89 |
+
"per_task": {
|
| 90 |
+
"LiftPegUpright-v1": {
|
| 91 |
+
"action_mse_to_best": 0.3295628765367509,
|
| 92 |
+
"expert_success_rate": 0.8865979381443299,
|
| 93 |
+
"num_groups": 97,
|
| 94 |
+
"oracle_success_rate": 0.9278350515463918,
|
| 95 |
+
"policy_expert_regret": 1.0783616742215205,
|
| 96 |
+
"policy_oracle_regret": 1.1047771519299634,
|
| 97 |
+
"policy_rollout_progress": 0.5714352193874183,
|
| 98 |
+
"policy_rollout_success_rate": 0.2268041237113402,
|
| 99 |
+
"restore_max_error": 2.384185791015625e-07
|
| 100 |
+
},
|
| 101 |
+
"PickCube-v1": {
|
| 102 |
+
"action_mse_to_best": 0.3043971854763535,
|
| 103 |
+
"expert_success_rate": 0.9375,
|
| 104 |
+
"num_groups": 208,
|
| 105 |
+
"oracle_success_rate": 0.9471153846153846,
|
| 106 |
+
"policy_expert_regret": 1.0234465546762714,
|
| 107 |
+
"policy_oracle_regret": 1.0368694221027768,
|
| 108 |
+
"policy_rollout_progress": 0.5922816276120452,
|
| 109 |
+
"policy_rollout_success_rate": 0.30288461538461536,
|
| 110 |
+
"restore_max_error": 2.384185791015625e-07
|
| 111 |
+
},
|
| 112 |
+
"PullCube-v1": {
|
| 113 |
+
"action_mse_to_best": 0.6128621583248114,
|
| 114 |
+
"expert_success_rate": 0.19480519480519481,
|
| 115 |
+
"num_groups": 77,
|
| 116 |
+
"oracle_success_rate": 0.36363636363636365,
|
| 117 |
+
"policy_expert_regret": 0.24367654795891472,
|
| 118 |
+
"policy_oracle_regret": 0.39216643712761534,
|
| 119 |
+
"policy_rollout_progress": 0.32449440497832444,
|
| 120 |
+
"policy_rollout_success_rate": 0.22077922077922077,
|
| 121 |
+
"restore_max_error": 2.3096799850463867e-07
|
| 122 |
+
},
|
| 123 |
+
"PushCube-v1": {
|
| 124 |
+
"action_mse_to_best": 0.36770542179967497,
|
| 125 |
+
"expert_success_rate": 0.8617021276595744,
|
| 126 |
+
"num_groups": 94,
|
| 127 |
+
"oracle_success_rate": 0.9893617021276596,
|
| 128 |
+
"policy_expert_regret": 0.5508309629686335,
|
| 129 |
+
"policy_oracle_regret": 0.628381609441118,
|
| 130 |
+
"policy_rollout_progress": 0.7014056246014352,
|
| 131 |
+
"policy_rollout_success_rate": 0.6702127659574468,
|
| 132 |
+
"restore_max_error": 2.384185791015625e-07
|
| 133 |
+
},
|
| 134 |
+
"StackCube-v1": {
|
| 135 |
+
"action_mse_to_best": 0.5510172197087244,
|
| 136 |
+
"expert_success_rate": 0.6767676767676768,
|
| 137 |
+
"num_groups": 99,
|
| 138 |
+
"oracle_success_rate": 0.8585858585858586,
|
| 139 |
+
"policy_expert_regret": 0.9308298939105236,
|
| 140 |
+
"policy_oracle_regret": 1.0785101525711291,
|
| 141 |
+
"policy_rollout_progress": 0.4711275805126537,
|
| 142 |
+
"policy_rollout_success_rate": 0.2828282828282828,
|
| 143 |
+
"restore_max_error": 2.384185791015625e-07
|
| 144 |
+
}
|
| 145 |
+
}
|
| 146 |
+
},
|
| 147 |
+
{
|
| 148 |
+
"seed": 1,
|
| 149 |
+
"path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/near_miss_policy_bc5/seed_1/policy_rollout_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p10_noopbonus0p03.json",
|
| 150 |
+
"num_groups": 575,
|
| 151 |
+
"selection_mode": "retrieval_residual",
|
| 152 |
+
"num_candidates": 48,
|
| 153 |
+
"candidate_sigma": 0.0,
|
| 154 |
+
"selection_margin": 0.1,
|
| 155 |
+
"prepend_policy_candidate": false,
|
| 156 |
+
"field_optim_steps": 0,
|
| 157 |
+
"field_optim_step_size": 0.0,
|
| 158 |
+
"field_optim_trust_radius": 0.0,
|
| 159 |
+
"field_optim_l2_penalty": 0.0,
|
| 160 |
+
"retrieval_neighbors": 4,
|
| 161 |
+
"retrieval_metric": "raw",
|
| 162 |
+
"retrieval_type_min_success": 0.0,
|
| 163 |
+
"retrieval_type_success_bonus_scale": 0.0,
|
| 164 |
+
"retrieval_residual_consensus_penalty_scale": 0.0,
|
| 165 |
+
"retrieval_residual_min_source_progress": 0.0,
|
| 166 |
+
"retrieval_residual_min_source_advantage": -1000000000.0,
|
| 167 |
+
"retrieval_residual_source_progress_bonus_scale": 0.0,
|
| 168 |
+
"retrieval_residual_source_score_bonus_scale": 0.0,
|
| 169 |
+
"retrieval_residual_source_advantage_bonus_scale": 0.0,
|
| 170 |
+
"retrieval_residual_composite_l2_penalty_scale": 0.0,
|
| 171 |
+
"retrieval_residual_scale": 1.0,
|
| 172 |
+
"retrieval_residual_scales": [
|
| 173 |
+
0.35,
|
| 174 |
+
0.4,
|
| 175 |
+
0.45
|
| 176 |
+
],
|
| 177 |
+
"retrieval_residual_anchor": "expert",
|
| 178 |
+
"retrieval_residual_direction": "candidate_minus_anchor",
|
| 179 |
+
"retrieval_residual_reduce": "compose_mean_by_type",
|
| 180 |
+
"lattice_exclude_types": [
|
| 181 |
+
"residual_random_negative",
|
| 182 |
+
"residual_wrong_direction",
|
| 183 |
+
"residual_near_miss+residual_no_op"
|
| 184 |
+
],
|
| 185 |
+
"candidate_type_bonuses": {
|
| 186 |
+
"residual_no_op": 0.03
|
| 187 |
+
},
|
| 188 |
+
"candidate_type_bonus_components": false,
|
| 189 |
+
"candidate_oracle_rollouts": 0,
|
| 190 |
+
"candidate_oracle_success_rate": null,
|
| 191 |
+
"candidate_oracle_progress": null,
|
| 192 |
+
"candidate_oracle_score": null,
|
| 193 |
+
"candidate_oracle_regret": null,
|
| 194 |
+
"candidate_oracle_score_gain_over_selected": null,
|
| 195 |
+
"candidate_oracle_improvement_rate": null,
|
| 196 |
+
"candidate_oracle_selected_branch_success_rate": null,
|
| 197 |
+
"candidate_oracle_selected_branch_progress": null,
|
| 198 |
+
"candidate_oracle_type_counts": {},
|
| 199 |
+
"selected_residual_scale_counts": {
|
| 200 |
+
"0.35": 504,
|
| 201 |
+
"0.45": 53,
|
| 202 |
+
"0.4": 18
|
| 203 |
+
},
|
| 204 |
+
"policy_rollout_success_rate": 0.35304347826086957,
|
| 205 |
+
"policy_rollout_progress": 0.573193692833833,
|
| 206 |
+
"oracle_success_rate": 0.8695652173913043,
|
| 207 |
+
"action_mse_to_best": 0.40915777628674455,
|
| 208 |
+
"best_policy_val": {
|
| 209 |
+
"bc_loss": 0.1315989200439718,
|
| 210 |
+
"field_effect_loss": 0.010110370814800262,
|
| 211 |
+
"field_potential_loss": 0.31546536915832096,
|
| 212 |
+
"field_preference_loss": 0.4960637440284093,
|
| 213 |
+
"lattice_edges": 3833.3333333333335,
|
| 214 |
+
"progress_mae": 0.19023226698239645,
|
| 215 |
+
"rank_acc": 0.8255469501018524,
|
| 216 |
+
"rank_loss": 0.4960531195004781,
|
| 217 |
+
"regret_mae": 0.42156532737943864,
|
| 218 |
+
"success_accuracy": 0.8807473679383596,
|
| 219 |
+
"total_loss": 1.561984618504842
|
| 220 |
+
},
|
| 221 |
+
"per_task": {
|
| 222 |
+
"LiftPegUpright-v1": {
|
| 223 |
+
"action_mse_to_best": 0.3299746959330102,
|
| 224 |
+
"expert_success_rate": 0.8584070796460177,
|
| 225 |
+
"num_groups": 113,
|
| 226 |
+
"oracle_success_rate": 0.9380530973451328,
|
| 227 |
+
"policy_expert_regret": 0.9873896409979964,
|
| 228 |
+
"policy_oracle_regret": 1.0665916710828258,
|
| 229 |
+
"policy_rollout_progress": 0.6355200547032651,
|
| 230 |
+
"policy_rollout_success_rate": 0.23008849557522124,
|
| 231 |
+
"restore_max_error": 2.384185791015625e-07
|
| 232 |
+
},
|
| 233 |
+
"PickCube-v1": {
|
| 234 |
+
"action_mse_to_best": 0.3384826354398999,
|
| 235 |
+
"expert_success_rate": 0.9402173913043478,
|
| 236 |
+
"num_groups": 184,
|
| 237 |
+
"oracle_success_rate": 0.9456521739130435,
|
| 238 |
+
"policy_expert_regret": 1.0418379528387247,
|
| 239 |
+
"policy_oracle_regret": 1.041861957313655,
|
| 240 |
+
"policy_rollout_progress": 0.5816247659512674,
|
| 241 |
+
"policy_rollout_success_rate": 0.31521739130434784,
|
| 242 |
+
"restore_max_error": 2.384185791015625e-07
|
| 243 |
+
},
|
| 244 |
+
"PullCube-v1": {
|
| 245 |
+
"action_mse_to_best": 0.6045245982118344,
|
| 246 |
+
"expert_success_rate": 0.25,
|
| 247 |
+
"num_groups": 76,
|
| 248 |
+
"oracle_success_rate": 0.40789473684210525,
|
| 249 |
+
"policy_expert_regret": 0.29825114321551827,
|
| 250 |
+
"policy_oracle_regret": 0.5216845717085036,
|
| 251 |
+
"policy_rollout_progress": 0.2906919057039838,
|
| 252 |
+
"policy_rollout_success_rate": 0.18421052631578946,
|
| 253 |
+
"restore_max_error": 2.384185791015625e-07
|
| 254 |
+
},
|
| 255 |
+
"PushCube-v1": {
|
| 256 |
+
"action_mse_to_best": 0.366618625215582,
|
| 257 |
+
"expert_success_rate": 0.8198198198198198,
|
| 258 |
+
"num_groups": 111,
|
| 259 |
+
"oracle_success_rate": 1.0,
|
| 260 |
+
"policy_expert_regret": 0.329537373002585,
|
| 261 |
+
"policy_oracle_regret": 0.34670494013541453,
|
| 262 |
+
"policy_rollout_progress": 0.8334752400447657,
|
| 263 |
+
"policy_rollout_success_rate": 0.8198198198198198,
|
| 264 |
+
"restore_max_error": 2.3795291781425476e-07
|
| 265 |
+
},
|
| 266 |
+
"StackCube-v1": {
|
| 267 |
+
"action_mse_to_best": 0.5391125158294217,
|
| 268 |
+
"expert_success_rate": 0.6923076923076923,
|
| 269 |
+
"num_groups": 91,
|
| 270 |
+
"oracle_success_rate": 0.8571428571428571,
|
| 271 |
+
"policy_expert_regret": 1.1337422296241089,
|
| 272 |
+
"policy_oracle_regret": 1.2934040767805917,
|
| 273 |
+
"policy_rollout_progress": 0.39720125037890214,
|
| 274 |
+
"policy_rollout_success_rate": 0.15384615384615385,
|
| 275 |
+
"restore_max_error": 1.9371509552001953e-07
|
| 276 |
+
}
|
| 277 |
+
}
|
| 278 |
+
},
|
| 279 |
+
{
|
| 280 |
+
"seed": 2,
|
| 281 |
+
"path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/near_miss_policy_bc5/seed_2/policy_rollout_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p10_noopbonus0p03.json",
|
| 282 |
+
"num_groups": 575,
|
| 283 |
+
"selection_mode": "retrieval_residual",
|
| 284 |
+
"num_candidates": 48,
|
| 285 |
+
"candidate_sigma": 0.0,
|
| 286 |
+
"selection_margin": 0.1,
|
| 287 |
+
"prepend_policy_candidate": false,
|
| 288 |
+
"field_optim_steps": 0,
|
| 289 |
+
"field_optim_step_size": 0.0,
|
| 290 |
+
"field_optim_trust_radius": 0.0,
|
| 291 |
+
"field_optim_l2_penalty": 0.0,
|
| 292 |
+
"retrieval_neighbors": 4,
|
| 293 |
+
"retrieval_metric": "raw",
|
| 294 |
+
"retrieval_type_min_success": 0.0,
|
| 295 |
+
"retrieval_type_success_bonus_scale": 0.0,
|
| 296 |
+
"retrieval_residual_consensus_penalty_scale": 0.0,
|
| 297 |
+
"retrieval_residual_min_source_progress": 0.0,
|
| 298 |
+
"retrieval_residual_min_source_advantage": -1000000000.0,
|
| 299 |
+
"retrieval_residual_source_progress_bonus_scale": 0.0,
|
| 300 |
+
"retrieval_residual_source_score_bonus_scale": 0.0,
|
| 301 |
+
"retrieval_residual_source_advantage_bonus_scale": 0.0,
|
| 302 |
+
"retrieval_residual_composite_l2_penalty_scale": 0.0,
|
| 303 |
+
"retrieval_residual_scale": 1.0,
|
| 304 |
+
"retrieval_residual_scales": [
|
| 305 |
+
0.35,
|
| 306 |
+
0.4,
|
| 307 |
+
0.45
|
| 308 |
+
],
|
| 309 |
+
"retrieval_residual_anchor": "expert",
|
| 310 |
+
"retrieval_residual_direction": "candidate_minus_anchor",
|
| 311 |
+
"retrieval_residual_reduce": "compose_mean_by_type",
|
| 312 |
+
"lattice_exclude_types": [
|
| 313 |
+
"residual_random_negative",
|
| 314 |
+
"residual_wrong_direction",
|
| 315 |
+
"residual_near_miss+residual_no_op"
|
| 316 |
+
],
|
| 317 |
+
"candidate_type_bonuses": {
|
| 318 |
+
"residual_no_op": 0.03
|
| 319 |
+
},
|
| 320 |
+
"candidate_type_bonus_components": false,
|
| 321 |
+
"candidate_oracle_rollouts": 0,
|
| 322 |
+
"candidate_oracle_success_rate": null,
|
| 323 |
+
"candidate_oracle_progress": null,
|
| 324 |
+
"candidate_oracle_score": null,
|
| 325 |
+
"candidate_oracle_regret": null,
|
| 326 |
+
"candidate_oracle_score_gain_over_selected": null,
|
| 327 |
+
"candidate_oracle_improvement_rate": null,
|
| 328 |
+
"candidate_oracle_selected_branch_success_rate": null,
|
| 329 |
+
"candidate_oracle_selected_branch_progress": null,
|
| 330 |
+
"candidate_oracle_type_counts": {},
|
| 331 |
+
"selected_residual_scale_counts": {
|
| 332 |
+
"0.35": 482,
|
| 333 |
+
"0.45": 71,
|
| 334 |
+
"0.4": 22
|
| 335 |
+
},
|
| 336 |
+
"policy_rollout_success_rate": 0.35130434782608694,
|
| 337 |
+
"policy_rollout_progress": 0.5703554131511761,
|
| 338 |
+
"oracle_success_rate": 0.8765217391304347,
|
| 339 |
+
"action_mse_to_best": 0.4414301192638991,
|
| 340 |
+
"best_policy_val": {
|
| 341 |
+
"bc_loss": 0.11367896075050037,
|
| 342 |
+
"field_effect_loss": 0.009670218582161598,
|
| 343 |
+
"field_potential_loss": 0.2641640139950646,
|
| 344 |
+
"field_preference_loss": 0.5130490180518892,
|
| 345 |
+
"lattice_edges": 3833.3333333333335,
|
| 346 |
+
"progress_mae": 0.2021729110015763,
|
| 347 |
+
"rank_acc": 0.8333857821093665,
|
| 348 |
+
"rank_loss": 0.5130119257503085,
|
| 349 |
+
"regret_mae": 0.3958987047274907,
|
| 350 |
+
"success_accuracy": 0.8680730561415354,
|
| 351 |
+
"total_loss": 1.4394984311527677
|
| 352 |
+
},
|
| 353 |
+
"per_task": {
|
| 354 |
+
"LiftPegUpright-v1": {
|
| 355 |
+
"action_mse_to_best": 0.3581597938561269,
|
| 356 |
+
"expert_success_rate": 0.8229166666666666,
|
| 357 |
+
"num_groups": 96,
|
| 358 |
+
"oracle_success_rate": 0.9270833333333334,
|
| 359 |
+
"policy_expert_regret": 0.8750739996321499,
|
| 360 |
+
"policy_oracle_regret": 0.9813436895298461,
|
| 361 |
+
"policy_rollout_progress": 0.6397648337297142,
|
| 362 |
+
"policy_rollout_success_rate": 0.3125,
|
| 363 |
+
"restore_max_error": 1.955777406692505e-07
|
| 364 |
+
},
|
| 365 |
+
"PickCube-v1": {
|
| 366 |
+
"action_mse_to_best": 0.36875359663231805,
|
| 367 |
+
"expert_success_rate": 0.9444444444444444,
|
| 368 |
+
"num_groups": 198,
|
| 369 |
+
"oracle_success_rate": 0.9595959595959596,
|
| 370 |
+
"policy_expert_regret": 1.009974913013102,
|
| 371 |
+
"policy_oracle_regret": 1.0242046085421486,
|
| 372 |
+
"policy_rollout_progress": 0.6195695759220556,
|
| 373 |
+
"policy_rollout_success_rate": 0.3181818181818182,
|
| 374 |
+
"restore_max_error": 2.384185791015625e-07
|
| 375 |
+
},
|
| 376 |
+
"PullCube-v1": {
|
| 377 |
+
"action_mse_to_best": 0.6482692137360573,
|
| 378 |
+
"expert_success_rate": 0.24444444444444444,
|
| 379 |
+
"num_groups": 90,
|
| 380 |
+
"oracle_success_rate": 0.4666666666666667,
|
| 381 |
+
"policy_expert_regret": 0.29343414877851803,
|
| 382 |
+
"policy_oracle_regret": 0.5605833363408844,
|
| 383 |
+
"policy_rollout_progress": 0.31182089023705983,
|
| 384 |
+
"policy_rollout_success_rate": 0.2,
|
| 385 |
+
"restore_max_error": 2.384185791015625e-07
|
| 386 |
+
},
|
| 387 |
+
"PushCube-v1": {
|
| 388 |
+
"action_mse_to_best": 0.40162252568372403,
|
| 389 |
+
"expert_success_rate": 0.8514851485148515,
|
| 390 |
+
"num_groups": 101,
|
| 391 |
+
"oracle_success_rate": 1.0,
|
| 392 |
+
"policy_expert_regret": 0.3614039024209032,
|
| 393 |
+
"policy_oracle_regret": 0.4350951622323235,
|
| 394 |
+
"policy_rollout_progress": 0.7926276100449042,
|
| 395 |
+
"policy_rollout_success_rate": 0.7722772277227723,
|
| 396 |
+
"restore_max_error": 2.384185791015625e-07
|
| 397 |
+
},
|
| 398 |
+
"StackCube-v1": {
|
| 399 |
+
"action_mse_to_best": 0.5279740211450391,
|
| 400 |
+
"expert_success_rate": 0.7666666666666667,
|
| 401 |
+
"num_groups": 90,
|
| 402 |
+
"oracle_success_rate": 0.9111111111111111,
|
| 403 |
+
"policy_expert_regret": 1.1824685187803374,
|
| 404 |
+
"policy_oracle_regret": 1.3445890701479382,
|
| 405 |
+
"policy_rollout_progress": 0.3971432639492883,
|
| 406 |
+
"policy_rollout_success_rate": 0.14444444444444443,
|
| 407 |
+
"restore_max_error": 2.2351741790771484e-07
|
| 408 |
+
}
|
| 409 |
+
}
|
| 410 |
+
}
|
| 411 |
+
]
|
| 412 |
+
}
|
results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p10_noopbonus0p03_summary.md
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# h=16 Best-Policy Checkpoint Rollout
|
| 2 |
+
|
| 3 |
+
Run root: `/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs`
|
| 4 |
+
Objective: `near_miss_policy_bc5`
|
| 5 |
+
Result file: `policy_rollout_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p10_noopbonus0p03.json`
|
| 6 |
+
Completed seeds: 3
|
| 7 |
+
Baseline h=4 policy success: 29.67%
|
| 8 |
+
Baseline h=16 rank-checkpoint success: 29.74%
|
| 9 |
+
|
| 10 |
+
Mean success: 34.67% +/- 0.96%
|
| 11 |
+
Gain vs h=16 rank checkpoint: +4.93%
|
| 12 |
+
Mean progress: 56.45%
|
| 13 |
+
Mean action MSE to best: 0.418
|
| 14 |
+
|
| 15 |
+
| seed | mode | k | policy cand | retrieval K | retrieval metric | residual anchor | residual direction | residual reduce | min type success | type success bonus | consensus penalty | min source progress | source progress bonus | source score bonus | residual scale | residual scales | margin | sigma | opt steps | trust | success | progress | oracle | candidate oracle | oracle gain | action MSE |
|
| 16 |
+
|---:|---|---:|---|---:|---|---|---|---|---:|---:|---:|---:|---:|---:|---:|---|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|
|
| 17 |
+
| 0 | retrieval_residual | 48 | no | 4 | raw | expert | candidate_minus_anchor | compose_mean_by_type | 0.00 | 0.000 | 0.000 | 0.00 | 0.000 | 0.000 | 1.00 | 0.35,0.40,0.45 | 0.100 | 0.00 | 0 | 0.00 | 33.57% | 54.99% | 85.74% | n/a | n/a | 0.403 |
|
| 18 |
+
| 1 | retrieval_residual | 48 | no | 4 | raw | expert | candidate_minus_anchor | compose_mean_by_type | 0.00 | 0.000 | 0.000 | 0.00 | 0.000 | 0.000 | 1.00 | 0.35,0.40,0.45 | 0.100 | 0.00 | 0 | 0.00 | 35.30% | 57.32% | 86.96% | n/a | n/a | 0.409 |
|
| 19 |
+
| 2 | retrieval_residual | 48 | no | 4 | raw | expert | candidate_minus_anchor | compose_mean_by_type | 0.00 | 0.000 | 0.000 | 0.00 | 0.000 | 0.000 | 1.00 | 0.35,0.40,0.45 | 0.100 | 0.00 | 0 | 0.00 | 35.13% | 57.04% | 87.65% | n/a | n/a | 0.441 |
|
results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_oraclek8_summary.json
ADDED
|
@@ -0,0 +1,697 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"run_root": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs",
|
| 3 |
+
"objective": "near_miss_policy_bc5",
|
| 4 |
+
"out_name": "policy_rollout_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_oraclek8.json",
|
| 5 |
+
"num_completed": 3,
|
| 6 |
+
"baseline_h4_policy_success": 0.2967,
|
| 7 |
+
"baseline_h16_rank_checkpoint_success": 0.29739130434782607,
|
| 8 |
+
"mean_success": 0.3565217391304348,
|
| 9 |
+
"std_success": 0.012173913043478276,
|
| 10 |
+
"mean_progress": 0.571575072284851,
|
| 11 |
+
"mean_action_mse_to_best": 0.4063519326314006,
|
| 12 |
+
"gain_vs_h4": 0.05982173913043476,
|
| 13 |
+
"gain_vs_h16_rank_checkpoint": 0.05913043478260871,
|
| 14 |
+
"rows": [
|
| 15 |
+
{
|
| 16 |
+
"seed": 0,
|
| 17 |
+
"path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/near_miss_policy_bc5/seed_0/policy_rollout_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_oraclek8.json",
|
| 18 |
+
"num_groups": 575,
|
| 19 |
+
"selection_mode": "retrieval_residual",
|
| 20 |
+
"num_candidates": 48,
|
| 21 |
+
"candidate_sigma": 0.0,
|
| 22 |
+
"selection_margin": 0.2,
|
| 23 |
+
"prepend_policy_candidate": false,
|
| 24 |
+
"field_optim_steps": 0,
|
| 25 |
+
"field_optim_step_size": 0.0,
|
| 26 |
+
"field_optim_trust_radius": 0.0,
|
| 27 |
+
"field_optim_l2_penalty": 0.0,
|
| 28 |
+
"retrieval_neighbors": 4,
|
| 29 |
+
"retrieval_metric": "raw",
|
| 30 |
+
"retrieval_type_min_success": 0.0,
|
| 31 |
+
"retrieval_type_success_bonus_scale": 0.0,
|
| 32 |
+
"retrieval_residual_consensus_penalty_scale": 0.0,
|
| 33 |
+
"retrieval_residual_min_source_progress": 0.0,
|
| 34 |
+
"retrieval_residual_min_source_advantage": -1000000000.0,
|
| 35 |
+
"retrieval_residual_source_progress_bonus_scale": 0.0,
|
| 36 |
+
"retrieval_residual_source_score_bonus_scale": 0.0,
|
| 37 |
+
"retrieval_residual_source_advantage_bonus_scale": 0.0,
|
| 38 |
+
"retrieval_residual_composite_l2_penalty_scale": 0.0,
|
| 39 |
+
"retrieval_residual_scale": 1.0,
|
| 40 |
+
"retrieval_residual_scales": [
|
| 41 |
+
0.35,
|
| 42 |
+
0.4,
|
| 43 |
+
0.45
|
| 44 |
+
],
|
| 45 |
+
"retrieval_residual_anchor": "expert",
|
| 46 |
+
"retrieval_residual_direction": "candidate_minus_anchor",
|
| 47 |
+
"retrieval_residual_reduce": "compose_mean_by_type",
|
| 48 |
+
"lattice_exclude_types": [
|
| 49 |
+
"residual_random_negative",
|
| 50 |
+
"residual_wrong_direction",
|
| 51 |
+
"residual_near_miss+residual_no_op"
|
| 52 |
+
],
|
| 53 |
+
"candidate_type_bonuses": {
|
| 54 |
+
"residual_no_op": 0.03
|
| 55 |
+
},
|
| 56 |
+
"candidate_type_bonus_components": false,
|
| 57 |
+
"candidate_oracle_rollouts": 8,
|
| 58 |
+
"candidate_oracle_success_rate": 0.41043478260869565,
|
| 59 |
+
"candidate_oracle_progress": 0.6247517613127179,
|
| 60 |
+
"candidate_oracle_score": 1.0351865439214136,
|
| 61 |
+
"candidate_oracle_regret": 0.7660745985449656,
|
| 62 |
+
"candidate_oracle_score_gain_over_selected": 0.1385079800337553,
|
| 63 |
+
"candidate_oracle_improvement_rate": 0.5391304347826087,
|
| 64 |
+
"candidate_oracle_selected_branch_success_rate": 0.3426086956521739,
|
| 65 |
+
"candidate_oracle_selected_branch_progress": 0.5540698683954289,
|
| 66 |
+
"candidate_oracle_type_counts": {
|
| 67 |
+
"retrieval_residual_policy_residual": 312,
|
| 68 |
+
"retrieval_residual_residual_near_miss": 75,
|
| 69 |
+
"retrieval_residual_residual_near_miss+residual_wrong_gripper": 20,
|
| 70 |
+
"retrieval_residual_residual_no_op": 58,
|
| 71 |
+
"retrieval_residual_residual_no_op+residual_wrong_gripper": 31,
|
| 72 |
+
"retrieval_residual_residual_wrong_gripper": 79
|
| 73 |
+
},
|
| 74 |
+
"selected_residual_scale_counts": {
|
| 75 |
+
"0.35": 536,
|
| 76 |
+
"0.45": 35,
|
| 77 |
+
"0.4": 4
|
| 78 |
+
},
|
| 79 |
+
"policy_rollout_success_rate": 0.35130434782608694,
|
| 80 |
+
"policy_rollout_progress": 0.5579317773906919,
|
| 81 |
+
"oracle_success_rate": 0.8573913043478261,
|
| 82 |
+
"action_mse_to_best": 0.3930192190424904,
|
| 83 |
+
"best_policy_val": {
|
| 84 |
+
"bc_loss": 0.13721593966086706,
|
| 85 |
+
"field_effect_loss": 0.009290305380192068,
|
| 86 |
+
"field_potential_loss": 0.2666468388504452,
|
| 87 |
+
"field_preference_loss": 0.5130573478009965,
|
| 88 |
+
"lattice_edges": 3833.3333333333335,
|
| 89 |
+
"progress_mae": 0.1933159919248687,
|
| 90 |
+
"rank_acc": 0.8265031774838766,
|
| 91 |
+
"rank_loss": 0.5130523675017886,
|
| 92 |
+
"regret_mae": 0.3756548762321472,
|
| 93 |
+
"success_accuracy": 0.8773836526605818,
|
| 94 |
+
"total_loss": 1.5581054819954767
|
| 95 |
+
},
|
| 96 |
+
"per_task": {
|
| 97 |
+
"LiftPegUpright-v1": {
|
| 98 |
+
"action_mse_to_best": 0.3206862742465335,
|
| 99 |
+
"candidate_oracle_improvement_rate": 0.5979381443298969,
|
| 100 |
+
"candidate_oracle_progress": 0.5828654941209813,
|
| 101 |
+
"candidate_oracle_regret": 1.0933468771964003,
|
| 102 |
+
"candidate_oracle_restore_max_error": 4.76837158203125e-07,
|
| 103 |
+
"candidate_oracle_score": 0.8096696178323215,
|
| 104 |
+
"candidate_oracle_score_gain_over_selected": 0.026710085494002115,
|
| 105 |
+
"candidate_oracle_selected_branch_progress": 0.5664646868238744,
|
| 106 |
+
"candidate_oracle_selected_branch_success_rate": 0.21649484536082475,
|
| 107 |
+
"candidate_oracle_success_rate": 0.2268041237113402,
|
| 108 |
+
"candidate_oracle_type_counts": {
|
| 109 |
+
"retrieval_residual_policy_residual": 69,
|
| 110 |
+
"retrieval_residual_residual_near_miss": 13,
|
| 111 |
+
"retrieval_residual_residual_near_miss+residual_wrong_gripper": 3,
|
| 112 |
+
"retrieval_residual_residual_no_op+residual_wrong_gripper": 1,
|
| 113 |
+
"retrieval_residual_residual_wrong_gripper": 11
|
| 114 |
+
},
|
| 115 |
+
"expert_success_rate": 0.8865979381443299,
|
| 116 |
+
"num_groups": 97,
|
| 117 |
+
"oracle_success_rate": 0.9278350515463918,
|
| 118 |
+
"policy_expert_regret": 1.078290365866779,
|
| 119 |
+
"policy_oracle_regret": 1.1046864339678557,
|
| 120 |
+
"policy_rollout_progress": 0.571525937349526,
|
| 121 |
+
"policy_rollout_success_rate": 0.2268041237113402,
|
| 122 |
+
"restore_max_error": 2.384185791015625e-07
|
| 123 |
+
},
|
| 124 |
+
"PickCube-v1": {
|
| 125 |
+
"action_mse_to_best": 0.30273104519941485,
|
| 126 |
+
"candidate_oracle_improvement_rate": 0.47596153846153844,
|
| 127 |
+
"candidate_oracle_progress": 0.6465907394706917,
|
| 128 |
+
"candidate_oracle_regret": 0.8960218487056688,
|
| 129 |
+
"candidate_oracle_restore_max_error": 4.76837158203125e-07,
|
| 130 |
+
"candidate_oracle_score": 1.0360138163937687,
|
| 131 |
+
"candidate_oracle_score_gain_over_selected": 0.06466827693825158,
|
| 132 |
+
"candidate_oracle_selected_branch_progress": 0.6203840004835421,
|
| 133 |
+
"candidate_oracle_selected_branch_success_rate": 0.35096153846153844,
|
| 134 |
+
"candidate_oracle_success_rate": 0.3894230769230769,
|
| 135 |
+
"candidate_oracle_type_counts": {
|
| 136 |
+
"retrieval_residual_policy_residual": 123,
|
| 137 |
+
"retrieval_residual_residual_near_miss": 26,
|
| 138 |
+
"retrieval_residual_residual_no_op": 44,
|
| 139 |
+
"retrieval_residual_residual_wrong_gripper": 15
|
| 140 |
+
},
|
| 141 |
+
"expert_success_rate": 0.9375,
|
| 142 |
+
"num_groups": 208,
|
| 143 |
+
"oracle_success_rate": 0.9471153846153846,
|
| 144 |
+
"policy_expert_regret": 0.9543257619505032,
|
| 145 |
+
"policy_oracle_regret": 0.9677486293770087,
|
| 146 |
+
"policy_rollout_progress": 0.6181331895685827,
|
| 147 |
+
"policy_rollout_success_rate": 0.34615384615384615,
|
| 148 |
+
"restore_max_error": 2.2910535335540771e-07
|
| 149 |
+
},
|
| 150 |
+
"PullCube-v1": {
|
| 151 |
+
"action_mse_to_best": 0.6033564122853341,
|
| 152 |
+
"candidate_oracle_improvement_rate": 0.7662337662337663,
|
| 153 |
+
"candidate_oracle_progress": 0.4401743694317418,
|
| 154 |
+
"candidate_oracle_regret": 0.26718359336547265,
|
| 155 |
+
"candidate_oracle_restore_max_error": 4.76837158203125e-07,
|
| 156 |
+
"candidate_oracle_score": 0.7648496941070665,
|
| 157 |
+
"candidate_oracle_score_gain_over_selected": 0.2697216790750042,
|
| 158 |
+
"candidate_oracle_selected_branch_progress": 0.30032282067136234,
|
| 159 |
+
"candidate_oracle_selected_branch_success_rate": 0.19480519480519481,
|
| 160 |
+
"candidate_oracle_success_rate": 0.3246753246753247,
|
| 161 |
+
"candidate_oracle_type_counts": {
|
| 162 |
+
"retrieval_residual_policy_residual": 20,
|
| 163 |
+
"retrieval_residual_residual_near_miss": 11,
|
| 164 |
+
"retrieval_residual_residual_near_miss+residual_wrong_gripper": 11,
|
| 165 |
+
"retrieval_residual_residual_no_op": 11,
|
| 166 |
+
"retrieval_residual_residual_no_op+residual_wrong_gripper": 6,
|
| 167 |
+
"retrieval_residual_residual_wrong_gripper": 18
|
| 168 |
+
},
|
| 169 |
+
"expert_success_rate": 0.19480519480519481,
|
| 170 |
+
"num_groups": 77,
|
| 171 |
+
"oracle_success_rate": 0.36363636363636365,
|
| 172 |
+
"policy_expert_regret": 0.2432419956728284,
|
| 173 |
+
"policy_oracle_regret": 0.41639802941459825,
|
| 174 |
+
"policy_rollout_progress": 0.3229174776722423,
|
| 175 |
+
"policy_rollout_success_rate": 0.22077922077922077,
|
| 176 |
+
"restore_max_error": 2.3096799850463867e-07
|
| 177 |
+
},
|
| 178 |
+
"PushCube-v1": {
|
| 179 |
+
"action_mse_to_best": 0.3675119048578942,
|
| 180 |
+
"candidate_oracle_improvement_rate": 0.3191489361702128,
|
| 181 |
+
"candidate_oracle_progress": 0.8114810194423858,
|
| 182 |
+
"candidate_oracle_regret": 0.3906466401320823,
|
| 183 |
+
"candidate_oracle_restore_max_error": 4.76837158203125e-07,
|
| 184 |
+
"candidate_oracle_score": 1.6093533598679177,
|
| 185 |
+
"candidate_oracle_score_gain_over_selected": 0.25869813735814806,
|
| 186 |
+
"candidate_oracle_selected_branch_progress": 0.6910807556928472,
|
| 187 |
+
"candidate_oracle_selected_branch_success_rate": 0.6595744680851063,
|
| 188 |
+
"candidate_oracle_success_rate": 0.7978723404255319,
|
| 189 |
+
"candidate_oracle_type_counts": {
|
| 190 |
+
"retrieval_residual_policy_residual": 64,
|
| 191 |
+
"retrieval_residual_residual_near_miss": 9,
|
| 192 |
+
"retrieval_residual_residual_near_miss+residual_wrong_gripper": 2,
|
| 193 |
+
"retrieval_residual_residual_no_op+residual_wrong_gripper": 2,
|
| 194 |
+
"retrieval_residual_residual_wrong_gripper": 17
|
| 195 |
+
},
|
| 196 |
+
"expert_success_rate": 0.8617021276595744,
|
| 197 |
+
"num_groups": 94,
|
| 198 |
+
"oracle_success_rate": 0.9893617021276596,
|
| 199 |
+
"policy_expert_regret": 0.5723979790159996,
|
| 200 |
+
"policy_oracle_regret": 0.6499486254884842,
|
| 201 |
+
"policy_rollout_progress": 0.6904769064264095,
|
| 202 |
+
"policy_rollout_success_rate": 0.6595744680851063,
|
| 203 |
+
"restore_max_error": 1.1920928955078125e-07
|
| 204 |
+
},
|
| 205 |
+
"StackCube-v1": {
|
| 206 |
+
"action_mse_to_best": 0.5142107287214862,
|
| 207 |
+
"candidate_oracle_improvement_rate": 0.6464646464646465,
|
| 208 |
+
"candidate_oracle_progress": 0.5861696337509636,
|
| 209 |
+
"candidate_oracle_regret": 0.9168866836663448,
|
| 210 |
+
"candidate_oracle_restore_max_error": 4.76837158203125e-07,
|
| 211 |
+
"candidate_oracle_score": 0.919502967084297,
|
| 212 |
+
"candidate_oracle_score_gain_over_selected": 0.1870102466958942,
|
| 213 |
+
"candidate_oracle_selected_branch_progress": 0.4698664583642073,
|
| 214 |
+
"candidate_oracle_selected_branch_success_rate": 0.26262626262626265,
|
| 215 |
+
"candidate_oracle_success_rate": 0.3333333333333333,
|
| 216 |
+
"candidate_oracle_type_counts": {
|
| 217 |
+
"retrieval_residual_policy_residual": 36,
|
| 218 |
+
"retrieval_residual_residual_near_miss": 16,
|
| 219 |
+
"retrieval_residual_residual_near_miss+residual_wrong_gripper": 4,
|
| 220 |
+
"retrieval_residual_residual_no_op": 3,
|
| 221 |
+
"retrieval_residual_residual_no_op+residual_wrong_gripper": 22,
|
| 222 |
+
"retrieval_residual_residual_wrong_gripper": 18
|
| 223 |
+
},
|
| 224 |
+
"expert_success_rate": 0.6767676767676768,
|
| 225 |
+
"num_groups": 99,
|
| 226 |
+
"oracle_success_rate": 0.8585858585858586,
|
| 227 |
+
"policy_expert_regret": 0.9052192963132955,
|
| 228 |
+
"policy_oracle_regret": 1.0644706299509665,
|
| 229 |
+
"policy_rollout_progress": 0.47506644102660095,
|
| 230 |
+
"policy_rollout_success_rate": 0.29292929292929293,
|
| 231 |
+
"restore_max_error": 2.1606683731079102e-07
|
| 232 |
+
}
|
| 233 |
+
}
|
| 234 |
+
},
|
| 235 |
+
{
|
| 236 |
+
"seed": 1,
|
| 237 |
+
"path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/near_miss_policy_bc5/seed_1/policy_rollout_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_oraclek8.json",
|
| 238 |
+
"num_groups": 575,
|
| 239 |
+
"selection_mode": "retrieval_residual",
|
| 240 |
+
"num_candidates": 48,
|
| 241 |
+
"candidate_sigma": 0.0,
|
| 242 |
+
"selection_margin": 0.2,
|
| 243 |
+
"prepend_policy_candidate": false,
|
| 244 |
+
"field_optim_steps": 0,
|
| 245 |
+
"field_optim_step_size": 0.0,
|
| 246 |
+
"field_optim_trust_radius": 0.0,
|
| 247 |
+
"field_optim_l2_penalty": 0.0,
|
| 248 |
+
"retrieval_neighbors": 4,
|
| 249 |
+
"retrieval_metric": "raw",
|
| 250 |
+
"retrieval_type_min_success": 0.0,
|
| 251 |
+
"retrieval_type_success_bonus_scale": 0.0,
|
| 252 |
+
"retrieval_residual_consensus_penalty_scale": 0.0,
|
| 253 |
+
"retrieval_residual_min_source_progress": 0.0,
|
| 254 |
+
"retrieval_residual_min_source_advantage": -1000000000.0,
|
| 255 |
+
"retrieval_residual_source_progress_bonus_scale": 0.0,
|
| 256 |
+
"retrieval_residual_source_score_bonus_scale": 0.0,
|
| 257 |
+
"retrieval_residual_source_advantage_bonus_scale": 0.0,
|
| 258 |
+
"retrieval_residual_composite_l2_penalty_scale": 0.0,
|
| 259 |
+
"retrieval_residual_scale": 1.0,
|
| 260 |
+
"retrieval_residual_scales": [
|
| 261 |
+
0.35,
|
| 262 |
+
0.4,
|
| 263 |
+
0.45
|
| 264 |
+
],
|
| 265 |
+
"retrieval_residual_anchor": "expert",
|
| 266 |
+
"retrieval_residual_direction": "candidate_minus_anchor",
|
| 267 |
+
"retrieval_residual_reduce": "compose_mean_by_type",
|
| 268 |
+
"lattice_exclude_types": [
|
| 269 |
+
"residual_random_negative",
|
| 270 |
+
"residual_wrong_direction",
|
| 271 |
+
"residual_near_miss+residual_no_op"
|
| 272 |
+
],
|
| 273 |
+
"candidate_type_bonuses": {
|
| 274 |
+
"residual_no_op": 0.03
|
| 275 |
+
},
|
| 276 |
+
"candidate_type_bonus_components": false,
|
| 277 |
+
"candidate_oracle_rollouts": 8,
|
| 278 |
+
"candidate_oracle_success_rate": 0.41043478260869565,
|
| 279 |
+
"candidate_oracle_progress": 0.6282550324855939,
|
| 280 |
+
"candidate_oracle_score": 1.0386898150942896,
|
| 281 |
+
"candidate_oracle_regret": 0.7745650228696025,
|
| 282 |
+
"candidate_oracle_score_gain_over_selected": 0.13500974421591863,
|
| 283 |
+
"candidate_oracle_improvement_rate": 0.5478260869565217,
|
| 284 |
+
"candidate_oracle_selected_branch_success_rate": 0.3391304347826087,
|
| 285 |
+
"candidate_oracle_selected_branch_progress": 0.5645496356908394,
|
| 286 |
+
"candidate_oracle_type_counts": {
|
| 287 |
+
"retrieval_residual_policy_residual": 302,
|
| 288 |
+
"retrieval_residual_residual_near_miss": 62,
|
| 289 |
+
"retrieval_residual_residual_near_miss+residual_wrong_gripper": 25,
|
| 290 |
+
"retrieval_residual_residual_no_op": 84,
|
| 291 |
+
"retrieval_residual_residual_no_op+residual_wrong_gripper": 21,
|
| 292 |
+
"retrieval_residual_residual_wrong_gripper": 81
|
| 293 |
+
},
|
| 294 |
+
"selected_residual_scale_counts": {
|
| 295 |
+
"0.35": 549,
|
| 296 |
+
"0.45": 23,
|
| 297 |
+
"0.4": 3
|
| 298 |
+
},
|
| 299 |
+
"policy_rollout_success_rate": 0.34782608695652173,
|
| 300 |
+
"policy_rollout_progress": 0.571580135051323,
|
| 301 |
+
"oracle_success_rate": 0.8695652173913043,
|
| 302 |
+
"action_mse_to_best": 0.39824521353549286,
|
| 303 |
+
"best_policy_val": {
|
| 304 |
+
"bc_loss": 0.1315989200439718,
|
| 305 |
+
"field_effect_loss": 0.010110370814800262,
|
| 306 |
+
"field_potential_loss": 0.31546536915832096,
|
| 307 |
+
"field_preference_loss": 0.4960637440284093,
|
| 308 |
+
"lattice_edges": 3833.3333333333335,
|
| 309 |
+
"progress_mae": 0.19023226698239645,
|
| 310 |
+
"rank_acc": 0.8255469501018524,
|
| 311 |
+
"rank_loss": 0.4960531195004781,
|
| 312 |
+
"regret_mae": 0.42156532737943864,
|
| 313 |
+
"success_accuracy": 0.8807473679383596,
|
| 314 |
+
"total_loss": 1.561984618504842
|
| 315 |
+
},
|
| 316 |
+
"per_task": {
|
| 317 |
+
"LiftPegUpright-v1": {
|
| 318 |
+
"action_mse_to_best": 0.3261080000104498,
|
| 319 |
+
"candidate_oracle_improvement_rate": 0.5663716814159292,
|
| 320 |
+
"candidate_oracle_progress": 0.6658792042099269,
|
| 321 |
+
"candidate_oracle_regret": 0.9742612073906755,
|
| 322 |
+
"candidate_oracle_restore_max_error": 4.76837158203125e-07,
|
| 323 |
+
"candidate_oracle_score": 0.9579146024400154,
|
| 324 |
+
"candidate_oracle_score_gain_over_selected": 0.11188626922337355,
|
| 325 |
+
"candidate_oracle_selected_branch_progress": 0.6247893938448577,
|
| 326 |
+
"candidate_oracle_selected_branch_success_rate": 0.22123893805309736,
|
| 327 |
+
"candidate_oracle_success_rate": 0.2920353982300885,
|
| 328 |
+
"candidate_oracle_type_counts": {
|
| 329 |
+
"retrieval_residual_policy_residual": 74,
|
| 330 |
+
"retrieval_residual_residual_near_miss": 17,
|
| 331 |
+
"retrieval_residual_residual_near_miss+residual_wrong_gripper": 5,
|
| 332 |
+
"retrieval_residual_residual_no_op": 3,
|
| 333 |
+
"retrieval_residual_residual_wrong_gripper": 14
|
| 334 |
+
},
|
| 335 |
+
"expert_success_rate": 0.8584070796460177,
|
| 336 |
+
"num_groups": 113,
|
| 337 |
+
"oracle_success_rate": 0.9380530973451328,
|
| 338 |
+
"policy_expert_regret": 0.9873958534635274,
|
| 339 |
+
"policy_oracle_regret": 1.0665980344061303,
|
| 340 |
+
"policy_rollout_progress": 0.6355136913799606,
|
| 341 |
+
"policy_rollout_success_rate": 0.23008849557522124,
|
| 342 |
+
"restore_max_error": 2.384185791015625e-07
|
| 343 |
+
},
|
| 344 |
+
"PickCube-v1": {
|
| 345 |
+
"action_mse_to_best": 0.32250052587250655,
|
| 346 |
+
"candidate_oracle_improvement_rate": 0.5760869565217391,
|
| 347 |
+
"candidate_oracle_progress": 0.605033606108602,
|
| 348 |
+
"candidate_oracle_regret": 0.9858444215041464,
|
| 349 |
+
"candidate_oracle_restore_max_error": 4.76837158203125e-07,
|
| 350 |
+
"candidate_oracle_score": 0.9528596930651237,
|
| 351 |
+
"candidate_oracle_score_gain_over_selected": 0.07707487480998364,
|
| 352 |
+
"candidate_oracle_selected_branch_progress": 0.5768717735722337,
|
| 353 |
+
"candidate_oracle_selected_branch_success_rate": 0.29891304347826086,
|
| 354 |
+
"candidate_oracle_success_rate": 0.34782608695652173,
|
| 355 |
+
"candidate_oracle_type_counts": {
|
| 356 |
+
"retrieval_residual_policy_residual": 90,
|
| 357 |
+
"retrieval_residual_residual_near_miss": 21,
|
| 358 |
+
"retrieval_residual_residual_no_op": 64,
|
| 359 |
+
"retrieval_residual_residual_no_op+residual_wrong_gripper": 2,
|
| 360 |
+
"retrieval_residual_residual_wrong_gripper": 7
|
| 361 |
+
},
|
| 362 |
+
"expert_success_rate": 0.9402173913043478,
|
| 363 |
+
"num_groups": 184,
|
| 364 |
+
"oracle_success_rate": 0.9456521739130435,
|
| 365 |
+
"policy_expert_regret": 1.0628609572152568,
|
| 366 |
+
"policy_oracle_regret": 1.0628849616901868,
|
| 367 |
+
"policy_rollout_progress": 0.5769061094008224,
|
| 368 |
+
"policy_rollout_success_rate": 0.29891304347826086,
|
| 369 |
+
"restore_max_error": 2.384185791015625e-07
|
| 370 |
+
},
|
| 371 |
+
"PullCube-v1": {
|
| 372 |
+
"action_mse_to_best": 0.5971616980080542,
|
| 373 |
+
"candidate_oracle_improvement_rate": 0.6973684210526315,
|
| 374 |
+
"candidate_oracle_progress": 0.38845373540626543,
|
| 375 |
+
"candidate_oracle_regret": 0.3710070996741323,
|
| 376 |
+
"candidate_oracle_restore_max_error": 4.76837158203125e-07,
|
| 377 |
+
"candidate_oracle_score": 0.6516116301431075,
|
| 378 |
+
"candidate_oracle_score_gain_over_selected": 0.20329608773126415,
|
| 379 |
+
"candidate_oracle_selected_branch_progress": 0.277262910881913,
|
| 380 |
+
"candidate_oracle_selected_branch_success_rate": 0.17105263157894737,
|
| 381 |
+
"candidate_oracle_success_rate": 0.2631578947368421,
|
| 382 |
+
"candidate_oracle_type_counts": {
|
| 383 |
+
"retrieval_residual_policy_residual": 27,
|
| 384 |
+
"retrieval_residual_residual_near_miss": 10,
|
| 385 |
+
"retrieval_residual_residual_near_miss+residual_wrong_gripper": 3,
|
| 386 |
+
"retrieval_residual_residual_no_op": 14,
|
| 387 |
+
"retrieval_residual_residual_no_op+residual_wrong_gripper": 6,
|
| 388 |
+
"retrieval_residual_residual_wrong_gripper": 16
|
| 389 |
+
},
|
| 390 |
+
"expert_success_rate": 0.25,
|
| 391 |
+
"num_groups": 76,
|
| 392 |
+
"oracle_success_rate": 0.40789473684210525,
|
| 393 |
+
"policy_expert_regret": 0.27669054590851855,
|
| 394 |
+
"policy_oracle_regret": 0.5001900389937586,
|
| 395 |
+
"policy_rollout_progress": 0.29926500726785316,
|
| 396 |
+
"policy_rollout_success_rate": 0.19736842105263158,
|
| 397 |
+
"restore_max_error": 2.3096799850463867e-07
|
| 398 |
+
},
|
| 399 |
+
"PushCube-v1": {
|
| 400 |
+
"action_mse_to_best": 0.3571035131677851,
|
| 401 |
+
"candidate_oracle_improvement_rate": 0.1981981981981982,
|
| 402 |
+
"candidate_oracle_progress": 0.9156151230002308,
|
| 403 |
+
"candidate_oracle_regret": 0.1744749670898592,
|
| 404 |
+
"candidate_oracle_restore_max_error": 4.76837158203125e-07,
|
| 405 |
+
"candidate_oracle_score": 1.8255250329101407,
|
| 406 |
+
"candidate_oracle_score_gain_over_selected": 0.20655315644569225,
|
| 407 |
+
"candidate_oracle_selected_branch_progress": 0.8171700761393383,
|
| 408 |
+
"candidate_oracle_selected_branch_success_rate": 0.8018018018018018,
|
| 409 |
+
"candidate_oracle_success_rate": 0.9099099099099099,
|
| 410 |
+
"candidate_oracle_type_counts": {
|
| 411 |
+
"retrieval_residual_policy_residual": 84,
|
| 412 |
+
"retrieval_residual_residual_near_miss": 4,
|
| 413 |
+
"retrieval_residual_residual_near_miss+residual_wrong_gripper": 3,
|
| 414 |
+
"retrieval_residual_residual_no_op": 1,
|
| 415 |
+
"retrieval_residual_residual_no_op+residual_wrong_gripper": 2,
|
| 416 |
+
"retrieval_residual_residual_wrong_gripper": 17
|
| 417 |
+
},
|
| 418 |
+
"expert_success_rate": 0.8198198198198198,
|
| 419 |
+
"num_groups": 111,
|
| 420 |
+
"oracle_success_rate": 1.0,
|
| 421 |
+
"policy_expert_regret": 0.3456981908362191,
|
| 422 |
+
"policy_oracle_regret": 0.36286575796904863,
|
| 423 |
+
"policy_rollout_progress": 0.8263234312201405,
|
| 424 |
+
"policy_rollout_success_rate": 0.8108108108108109,
|
| 425 |
+
"restore_max_error": 1.7881393432617188e-07
|
| 426 |
+
},
|
| 427 |
+
"StackCube-v1": {
|
| 428 |
+
"action_mse_to_best": 0.5250320660543966,
|
| 429 |
+
"candidate_oracle_improvement_rate": 0.7692307692307693,
|
| 430 |
+
"candidate_oracle_progress": 0.4782455773143978,
|
| 431 |
+
"candidate_oracle_regret": 1.1684037058890522,
|
| 432 |
+
"candidate_oracle_restore_max_error": 4.76837158203125e-07,
|
| 433 |
+
"candidate_oracle_score": 0.6760477751165956,
|
| 434 |
+
"candidate_oracle_score_gain_over_selected": 0.13656895059150653,
|
| 435 |
+
"candidate_oracle_selected_branch_progress": 0.3966216813404482,
|
| 436 |
+
"candidate_oracle_selected_branch_success_rate": 0.14285714285714285,
|
| 437 |
+
"candidate_oracle_success_rate": 0.1978021978021978,
|
| 438 |
+
"candidate_oracle_type_counts": {
|
| 439 |
+
"retrieval_residual_policy_residual": 27,
|
| 440 |
+
"retrieval_residual_residual_near_miss": 10,
|
| 441 |
+
"retrieval_residual_residual_near_miss+residual_wrong_gripper": 14,
|
| 442 |
+
"retrieval_residual_residual_no_op": 2,
|
| 443 |
+
"retrieval_residual_residual_no_op+residual_wrong_gripper": 11,
|
| 444 |
+
"retrieval_residual_residual_wrong_gripper": 27
|
| 445 |
+
},
|
| 446 |
+
"expert_success_rate": 0.6923076923076923,
|
| 447 |
+
"num_groups": 91,
|
| 448 |
+
"oracle_success_rate": 0.8571428571428571,
|
| 449 |
+
"policy_expert_regret": 1.1328262393618678,
|
| 450 |
+
"policy_oracle_regret": 1.292487030664643,
|
| 451 |
+
"policy_rollout_progress": 0.39811829649485075,
|
| 452 |
+
"policy_rollout_success_rate": 0.15384615384615385,
|
| 453 |
+
"restore_max_error": 1.9371509552001953e-07
|
| 454 |
+
}
|
| 455 |
+
}
|
| 456 |
+
},
|
| 457 |
+
{
|
| 458 |
+
"seed": 2,
|
| 459 |
+
"path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/near_miss_policy_bc5/seed_2/policy_rollout_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_oraclek8.json",
|
| 460 |
+
"num_groups": 575,
|
| 461 |
+
"selection_mode": "retrieval_residual",
|
| 462 |
+
"num_candidates": 48,
|
| 463 |
+
"candidate_sigma": 0.0,
|
| 464 |
+
"selection_margin": 0.2,
|
| 465 |
+
"prepend_policy_candidate": false,
|
| 466 |
+
"field_optim_steps": 0,
|
| 467 |
+
"field_optim_step_size": 0.0,
|
| 468 |
+
"field_optim_trust_radius": 0.0,
|
| 469 |
+
"field_optim_l2_penalty": 0.0,
|
| 470 |
+
"retrieval_neighbors": 4,
|
| 471 |
+
"retrieval_metric": "raw",
|
| 472 |
+
"retrieval_type_min_success": 0.0,
|
| 473 |
+
"retrieval_type_success_bonus_scale": 0.0,
|
| 474 |
+
"retrieval_residual_consensus_penalty_scale": 0.0,
|
| 475 |
+
"retrieval_residual_min_source_progress": 0.0,
|
| 476 |
+
"retrieval_residual_min_source_advantage": -1000000000.0,
|
| 477 |
+
"retrieval_residual_source_progress_bonus_scale": 0.0,
|
| 478 |
+
"retrieval_residual_source_score_bonus_scale": 0.0,
|
| 479 |
+
"retrieval_residual_source_advantage_bonus_scale": 0.0,
|
| 480 |
+
"retrieval_residual_composite_l2_penalty_scale": 0.0,
|
| 481 |
+
"retrieval_residual_scale": 1.0,
|
| 482 |
+
"retrieval_residual_scales": [
|
| 483 |
+
0.35,
|
| 484 |
+
0.4,
|
| 485 |
+
0.45
|
| 486 |
+
],
|
| 487 |
+
"retrieval_residual_anchor": "expert",
|
| 488 |
+
"retrieval_residual_direction": "candidate_minus_anchor",
|
| 489 |
+
"retrieval_residual_reduce": "compose_mean_by_type",
|
| 490 |
+
"lattice_exclude_types": [
|
| 491 |
+
"residual_random_negative",
|
| 492 |
+
"residual_wrong_direction",
|
| 493 |
+
"residual_near_miss+residual_no_op"
|
| 494 |
+
],
|
| 495 |
+
"candidate_type_bonuses": {
|
| 496 |
+
"residual_no_op": 0.03
|
| 497 |
+
},
|
| 498 |
+
"candidate_type_bonus_components": false,
|
| 499 |
+
"candidate_oracle_rollouts": 8,
|
| 500 |
+
"candidate_oracle_success_rate": 0.44521739130434784,
|
| 501 |
+
"candidate_oracle_progress": 0.6561966300010681,
|
| 502 |
+
"candidate_oracle_score": 1.101414021305416,
|
| 503 |
+
"candidate_oracle_regret": 0.7211246234178543,
|
| 504 |
+
"candidate_oracle_score_gain_over_selected": 0.1610163029672011,
|
| 505 |
+
"candidate_oracle_improvement_rate": 0.5165217391304348,
|
| 506 |
+
"candidate_oracle_selected_branch_success_rate": 0.3617391304347826,
|
| 507 |
+
"candidate_oracle_selected_branch_progress": 0.5786585876064894,
|
| 508 |
+
"candidate_oracle_type_counts": {
|
| 509 |
+
"retrieval_residual_policy_residual": 331,
|
| 510 |
+
"retrieval_residual_residual_near_miss": 74,
|
| 511 |
+
"retrieval_residual_residual_near_miss+residual_wrong_gripper": 31,
|
| 512 |
+
"retrieval_residual_residual_no_op": 38,
|
| 513 |
+
"retrieval_residual_residual_no_op+residual_wrong_gripper": 25,
|
| 514 |
+
"retrieval_residual_residual_wrong_gripper": 76
|
| 515 |
+
},
|
| 516 |
+
"selected_residual_scale_counts": {
|
| 517 |
+
"0.35": 546,
|
| 518 |
+
"0.45": 25,
|
| 519 |
+
"0.4": 4
|
| 520 |
+
},
|
| 521 |
+
"policy_rollout_success_rate": 0.37043478260869567,
|
| 522 |
+
"policy_rollout_progress": 0.5852133044125383,
|
| 523 |
+
"oracle_success_rate": 0.8765217391304347,
|
| 524 |
+
"action_mse_to_best": 0.42779136531621864,
|
| 525 |
+
"best_policy_val": {
|
| 526 |
+
"bc_loss": 0.11367896075050037,
|
| 527 |
+
"field_effect_loss": 0.009670218582161598,
|
| 528 |
+
"field_potential_loss": 0.2641640139950646,
|
| 529 |
+
"field_preference_loss": 0.5130490180518892,
|
| 530 |
+
"lattice_edges": 3833.3333333333335,
|
| 531 |
+
"progress_mae": 0.2021729110015763,
|
| 532 |
+
"rank_acc": 0.8333857821093665,
|
| 533 |
+
"rank_loss": 0.5130119257503085,
|
| 534 |
+
"regret_mae": 0.3958987047274907,
|
| 535 |
+
"success_accuracy": 0.8680730561415354,
|
| 536 |
+
"total_loss": 1.4394984311527677
|
| 537 |
+
},
|
| 538 |
+
"per_task": {
|
| 539 |
+
"LiftPegUpright-v1": {
|
| 540 |
+
"action_mse_to_best": 0.35442226944239036,
|
| 541 |
+
"candidate_oracle_improvement_rate": 0.5416666666666666,
|
| 542 |
+
"candidate_oracle_progress": 0.665017114020884,
|
| 543 |
+
"candidate_oracle_regret": 0.8935914092386762,
|
| 544 |
+
"candidate_oracle_restore_max_error": 4.76837158203125e-07,
|
| 545 |
+
"candidate_oracle_score": 1.040017114020884,
|
| 546 |
+
"candidate_oracle_score_gain_over_selected": 0.10511643424009283,
|
| 547 |
+
"candidate_oracle_selected_branch_progress": 0.6328173445848128,
|
| 548 |
+
"candidate_oracle_selected_branch_success_rate": 0.3020833333333333,
|
| 549 |
+
"candidate_oracle_success_rate": 0.375,
|
| 550 |
+
"candidate_oracle_type_counts": {
|
| 551 |
+
"retrieval_residual_policy_residual": 59,
|
| 552 |
+
"retrieval_residual_residual_near_miss": 15,
|
| 553 |
+
"retrieval_residual_residual_near_miss+residual_wrong_gripper": 2,
|
| 554 |
+
"retrieval_residual_residual_no_op+residual_wrong_gripper": 1,
|
| 555 |
+
"retrieval_residual_residual_wrong_gripper": 19
|
| 556 |
+
},
|
| 557 |
+
"expert_success_rate": 0.8229166666666666,
|
| 558 |
+
"num_groups": 96,
|
| 559 |
+
"oracle_success_rate": 0.9270833333333334,
|
| 560 |
+
"policy_expert_regret": 0.8475987246880928,
|
| 561 |
+
"policy_oracle_regret": 0.9538775502393643,
|
| 562 |
+
"policy_rollout_progress": 0.6463976396868626,
|
| 563 |
+
"policy_rollout_success_rate": 0.3333333333333333,
|
| 564 |
+
"restore_max_error": 2.384185791015625e-07
|
| 565 |
+
},
|
| 566 |
+
"PickCube-v1": {
|
| 567 |
+
"action_mse_to_best": 0.3390917487705898,
|
| 568 |
+
"candidate_oracle_improvement_rate": 0.47474747474747475,
|
| 569 |
+
"candidate_oracle_progress": 0.6511383742903067,
|
| 570 |
+
"candidate_oracle_regret": 0.942130759668847,
|
| 571 |
+
"candidate_oracle_restore_max_error": 4.76837158203125e-07,
|
| 572 |
+
"candidate_oracle_score": 1.0198252429771755,
|
| 573 |
+
"candidate_oracle_score_gain_over_selected": 0.045726843244123336,
|
| 574 |
+
"candidate_oracle_selected_branch_progress": 0.6357145610387729,
|
| 575 |
+
"candidate_oracle_selected_branch_success_rate": 0.3383838383838384,
|
| 576 |
+
"candidate_oracle_success_rate": 0.3686868686868687,
|
| 577 |
+
"candidate_oracle_type_counts": {
|
| 578 |
+
"retrieval_residual_policy_residual": 133,
|
| 579 |
+
"retrieval_residual_residual_near_miss": 25,
|
| 580 |
+
"retrieval_residual_residual_near_miss+residual_wrong_gripper": 3,
|
| 581 |
+
"retrieval_residual_residual_no_op": 26,
|
| 582 |
+
"retrieval_residual_residual_no_op+residual_wrong_gripper": 3,
|
| 583 |
+
"retrieval_residual_residual_wrong_gripper": 8
|
| 584 |
+
},
|
| 585 |
+
"expert_success_rate": 0.9444444444444444,
|
| 586 |
+
"num_groups": 198,
|
| 587 |
+
"oracle_success_rate": 0.9595959595959596,
|
| 588 |
+
"policy_expert_regret": 0.9807323978442435,
|
| 589 |
+
"policy_oracle_regret": 0.9878158483729519,
|
| 590 |
+
"policy_rollout_progress": 0.6357563158892321,
|
| 591 |
+
"policy_rollout_success_rate": 0.3383838383838384,
|
| 592 |
+
"restore_max_error": 2.384185791015625e-07
|
| 593 |
+
},
|
| 594 |
+
"PullCube-v1": {
|
| 595 |
+
"action_mse_to_best": 0.6408373928732343,
|
| 596 |
+
"candidate_oracle_improvement_rate": 0.6777777777777778,
|
| 597 |
+
"candidate_oracle_progress": 0.46258504376229315,
|
| 598 |
+
"candidate_oracle_regret": 0.30522582164655127,
|
| 599 |
+
"candidate_oracle_restore_max_error": 4.76837158203125e-07,
|
| 600 |
+
"candidate_oracle_score": 0.8292517104289598,
|
| 601 |
+
"candidate_oracle_score_gain_over_selected": 0.28015006221830846,
|
| 602 |
+
"candidate_oracle_selected_branch_progress": 0.326879425933243,
|
| 603 |
+
"candidate_oracle_selected_branch_success_rate": 0.2222222222222222,
|
| 604 |
+
"candidate_oracle_success_rate": 0.36666666666666664,
|
| 605 |
+
"candidate_oracle_type_counts": {
|
| 606 |
+
"retrieval_residual_policy_residual": 32,
|
| 607 |
+
"retrieval_residual_residual_near_miss": 19,
|
| 608 |
+
"retrieval_residual_residual_near_miss+residual_wrong_gripper": 18,
|
| 609 |
+
"retrieval_residual_residual_no_op": 7,
|
| 610 |
+
"retrieval_residual_residual_no_op+residual_wrong_gripper": 9,
|
| 611 |
+
"retrieval_residual_residual_wrong_gripper": 5
|
| 612 |
+
},
|
| 613 |
+
"expert_success_rate": 0.24444444444444444,
|
| 614 |
+
"num_groups": 90,
|
| 615 |
+
"oracle_success_rate": 0.4666666666666667,
|
| 616 |
+
"policy_expert_regret": 0.27485709413195863,
|
| 617 |
+
"policy_oracle_regret": 0.5026470268076183,
|
| 618 |
+
"policy_rollout_progress": 0.3366318846604372,
|
| 619 |
+
"policy_rollout_success_rate": 0.23333333333333334,
|
| 620 |
+
"restore_max_error": 2.384185791015625e-07
|
| 621 |
+
},
|
| 622 |
+
"PushCube-v1": {
|
| 623 |
+
"action_mse_to_best": 0.3970492488824495,
|
| 624 |
+
"candidate_oracle_improvement_rate": 0.2079207920792079,
|
| 625 |
+
"candidate_oracle_progress": 0.9267168732562868,
|
| 626 |
+
"candidate_oracle_regret": 0.15249104753579243,
|
| 627 |
+
"candidate_oracle_restore_max_error": 4.76837158203125e-07,
|
| 628 |
+
"candidate_oracle_score": 1.8475089524642077,
|
| 629 |
+
"candidate_oracle_score_gain_over_selected": 0.24635549538796492,
|
| 630 |
+
"candidate_oracle_selected_branch_progress": 0.8090742473850155,
|
| 631 |
+
"candidate_oracle_selected_branch_success_rate": 0.7920792079207921,
|
| 632 |
+
"candidate_oracle_success_rate": 0.9207920792079208,
|
| 633 |
+
"candidate_oracle_type_counts": {
|
| 634 |
+
"retrieval_residual_policy_residual": 74,
|
| 635 |
+
"retrieval_residual_residual_near_miss": 7,
|
| 636 |
+
"retrieval_residual_residual_near_miss+residual_wrong_gripper": 1,
|
| 637 |
+
"retrieval_residual_residual_no_op+residual_wrong_gripper": 2,
|
| 638 |
+
"retrieval_residual_residual_wrong_gripper": 17
|
| 639 |
+
},
|
| 640 |
+
"expert_success_rate": 0.8514851485148515,
|
| 641 |
+
"num_groups": 101,
|
| 642 |
+
"oracle_success_rate": 1.0,
|
| 643 |
+
"policy_expert_regret": 0.3425388766987489,
|
| 644 |
+
"policy_oracle_regret": 0.39744112810285964,
|
| 645 |
+
"policy_rollout_progress": 0.8104796639763483,
|
| 646 |
+
"policy_rollout_success_rate": 0.7920792079207921,
|
| 647 |
+
"restore_max_error": 2.384185791015625e-07
|
| 648 |
+
},
|
| 649 |
+
"StackCube-v1": {
|
| 650 |
+
"action_mse_to_best": 0.5226443493117888,
|
| 651 |
+
"candidate_oracle_improvement_rate": 0.7666666666666667,
|
| 652 |
+
"candidate_oracle_progress": 0.5479440339737468,
|
| 653 |
+
"candidate_oracle_regret": 1.1049785890512995,
|
| 654 |
+
"candidate_oracle_restore_max_error": 4.76837158203125e-07,
|
| 655 |
+
"candidate_oracle_score": 0.7812773673070802,
|
| 656 |
+
"candidate_oracle_score_gain_over_selected": 0.2593763435880343,
|
| 657 |
+
"candidate_oracle_selected_branch_progress": 0.3885676932003763,
|
| 658 |
+
"candidate_oracle_selected_branch_success_rate": 0.13333333333333333,
|
| 659 |
+
"candidate_oracle_success_rate": 0.23333333333333334,
|
| 660 |
+
"candidate_oracle_type_counts": {
|
| 661 |
+
"retrieval_residual_policy_residual": 33,
|
| 662 |
+
"retrieval_residual_residual_near_miss": 8,
|
| 663 |
+
"retrieval_residual_residual_near_miss+residual_wrong_gripper": 7,
|
| 664 |
+
"retrieval_residual_residual_no_op": 5,
|
| 665 |
+
"retrieval_residual_residual_no_op+residual_wrong_gripper": 10,
|
| 666 |
+
"retrieval_residual_residual_wrong_gripper": 27
|
| 667 |
+
},
|
| 668 |
+
"expert_success_rate": 0.7666666666666667,
|
| 669 |
+
"num_groups": 90,
|
| 670 |
+
"oracle_success_rate": 0.9111111111111111,
|
| 671 |
+
"policy_expert_regret": 1.1750739713509877,
|
| 672 |
+
"policy_oracle_regret": 1.3371944083107843,
|
| 673 |
+
"policy_rollout_progress": 0.4045378933350245,
|
| 674 |
+
"policy_rollout_success_rate": 0.14444444444444443,
|
| 675 |
+
"restore_max_error": 2.086162567138672e-07
|
| 676 |
+
}
|
| 677 |
+
}
|
| 678 |
+
}
|
| 679 |
+
],
|
| 680 |
+
"candidate_oracle_rollouts": 8,
|
| 681 |
+
"mean_candidate_oracle_success_rate": 0.4220289855072464,
|
| 682 |
+
"mean_candidate_oracle_progress": 0.63640114126646,
|
| 683 |
+
"mean_candidate_oracle_score": 1.0584301267737064,
|
| 684 |
+
"mean_candidate_oracle_regret": 0.7539214149441408,
|
| 685 |
+
"mean_candidate_oracle_score_gain_over_selected": 0.14484467573895835,
|
| 686 |
+
"mean_candidate_oracle_improvement_rate": 0.5344927536231884,
|
| 687 |
+
"mean_candidate_oracle_selected_branch_success_rate": 0.34782608695652173,
|
| 688 |
+
"mean_candidate_oracle_selected_branch_progress": 0.5657593638975859,
|
| 689 |
+
"candidate_oracle_type_counts": {
|
| 690 |
+
"retrieval_residual_policy_residual": 945,
|
| 691 |
+
"retrieval_residual_residual_near_miss": 211,
|
| 692 |
+
"retrieval_residual_residual_near_miss+residual_wrong_gripper": 76,
|
| 693 |
+
"retrieval_residual_residual_no_op": 180,
|
| 694 |
+
"retrieval_residual_residual_no_op+residual_wrong_gripper": 77,
|
| 695 |
+
"retrieval_residual_residual_wrong_gripper": 236
|
| 696 |
+
}
|
| 697 |
+
}
|
results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_oraclek8_summary.md
ADDED
|
@@ -0,0 +1,24 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# h=16 Best-Policy Checkpoint Rollout
|
| 2 |
+
|
| 3 |
+
Run root: `/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs`
|
| 4 |
+
Objective: `near_miss_policy_bc5`
|
| 5 |
+
Result file: `policy_rollout_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_oraclek8.json`
|
| 6 |
+
Completed seeds: 3
|
| 7 |
+
Baseline h=4 policy success: 29.67%
|
| 8 |
+
Baseline h=16 rank-checkpoint success: 29.74%
|
| 9 |
+
|
| 10 |
+
Mean success: 35.65% +/- 1.22%
|
| 11 |
+
Gain vs h=16 rank checkpoint: +5.91%
|
| 12 |
+
Mean progress: 57.16%
|
| 13 |
+
Mean action MSE to best: 0.406
|
| 14 |
+
Candidate-oracle prefix: top 8 candidates
|
| 15 |
+
Candidate-oracle success: 42.20%
|
| 16 |
+
Candidate-oracle progress: 63.64%
|
| 17 |
+
Candidate-oracle gain over selected branch: +0.145
|
| 18 |
+
Candidate-oracle improvement rate: 53.45%
|
| 19 |
+
|
| 20 |
+
| seed | mode | k | policy cand | retrieval K | retrieval metric | residual anchor | residual direction | residual reduce | min type success | type success bonus | consensus penalty | min source progress | source progress bonus | source score bonus | residual scale | residual scales | margin | sigma | opt steps | trust | success | progress | oracle | candidate oracle | oracle gain | action MSE |
|
| 21 |
+
|---:|---|---:|---|---:|---|---|---|---|---:|---:|---:|---:|---:|---:|---:|---|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|
|
| 22 |
+
| 0 | retrieval_residual | 48 | no | 4 | raw | expert | candidate_minus_anchor | compose_mean_by_type | 0.00 | 0.000 | 0.000 | 0.00 | 0.000 | 0.000 | 1.00 | 0.35,0.40,0.45 | 0.200 | 0.00 | 0 | 0.00 | 35.13% | 55.79% | 85.74% | 41.04% | +0.139 | 0.393 |
|
| 23 |
+
| 1 | retrieval_residual | 48 | no | 4 | raw | expert | candidate_minus_anchor | compose_mean_by_type | 0.00 | 0.000 | 0.000 | 0.00 | 0.000 | 0.000 | 1.00 | 0.35,0.40,0.45 | 0.200 | 0.00 | 0 | 0.00 | 34.78% | 57.16% | 86.96% | 41.04% | +0.135 | 0.398 |
|
| 24 |
+
| 2 | retrieval_residual | 48 | no | 4 | raw | expert | candidate_minus_anchor | compose_mean_by_type | 0.00 | 0.000 | 0.000 | 0.00 | 0.000 | 0.000 | 1.00 | 0.35,0.40,0.45 | 0.200 | 0.00 | 0 | 0.00 | 37.04% | 58.52% | 87.65% | 44.52% | +0.161 | 0.428 |
|