Auto-sync: 2026-06-28 08:14:39
Browse files- results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_scale0p35_safe_types_margin0p30_summary.json +292 -0
- results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_scale0p35_safe_types_margin0p30_summary.md +19 -0
- results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_scale0p35_safe_types_margin0p40_summary.json +292 -0
- results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_scale0p35_safe_types_margin0p40_summary.md +19 -0
- results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_scale0p35_safe_types_margin0p60_summary.json +292 -0
- results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_scale0p35_safe_types_margin0p60_summary.md +19 -0
- results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_scale0p35_safe_types_margin1p00_summary.json +292 -0
- results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_scale0p35_safe_types_margin1p00_summary.md +19 -0
results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_scale0p35_safe_types_margin0p30_summary.json
ADDED
|
@@ -0,0 +1,292 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 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_scale0p35_safe_types_margin0p30.json",
|
| 5 |
+
"num_completed": 3,
|
| 6 |
+
"baseline_h4_policy_success": 0.2967,
|
| 7 |
+
"baseline_h16_rank_checkpoint_success": 0.29739130434782607,
|
| 8 |
+
"mean_success": 0.34608695652173915,
|
| 9 |
+
"std_success": 0.018157054798105306,
|
| 10 |
+
"mean_progress": 0.5636367717266968,
|
| 11 |
+
"mean_action_mse_to_best": 0.39438031803870544,
|
| 12 |
+
"gain_vs_h4": 0.049386956521739134,
|
| 13 |
+
"gain_vs_h16_rank_checkpoint": 0.048695652173913084,
|
| 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_scale0p35_safe_types_margin0p30.json",
|
| 18 |
+
"num_groups": 575,
|
| 19 |
+
"selection_mode": "retrieval_residual",
|
| 20 |
+
"num_candidates": 16,
|
| 21 |
+
"candidate_sigma": 0.0,
|
| 22 |
+
"selection_margin": 0.3,
|
| 23 |
+
"field_optim_steps": 0,
|
| 24 |
+
"field_optim_step_size": 0.0,
|
| 25 |
+
"field_optim_trust_radius": 0.0,
|
| 26 |
+
"field_optim_l2_penalty": 0.0,
|
| 27 |
+
"retrieval_neighbors": 1,
|
| 28 |
+
"retrieval_metric": "raw",
|
| 29 |
+
"retrieval_type_min_success": 0.0,
|
| 30 |
+
"retrieval_residual_scale": 0.35,
|
| 31 |
+
"retrieval_residual_anchor": "expert",
|
| 32 |
+
"policy_rollout_success_rate": 0.3373913043478261,
|
| 33 |
+
"policy_rollout_progress": 0.5497364365742744,
|
| 34 |
+
"oracle_success_rate": 0.8573913043478261,
|
| 35 |
+
"action_mse_to_best": 0.37981279005094065,
|
| 36 |
+
"best_policy_val": {
|
| 37 |
+
"bc_loss": 0.13721593966086706,
|
| 38 |
+
"field_effect_loss": 0.009290305380192068,
|
| 39 |
+
"field_potential_loss": 0.2666468388504452,
|
| 40 |
+
"field_preference_loss": 0.5130573478009965,
|
| 41 |
+
"lattice_edges": 3833.3333333333335,
|
| 42 |
+
"progress_mae": 0.1933159919248687,
|
| 43 |
+
"rank_acc": 0.8265031774838766,
|
| 44 |
+
"rank_loss": 0.5130523675017886,
|
| 45 |
+
"regret_mae": 0.3756548762321472,
|
| 46 |
+
"success_accuracy": 0.8773836526605818,
|
| 47 |
+
"total_loss": 1.5581054819954767
|
| 48 |
+
},
|
| 49 |
+
"per_task": {
|
| 50 |
+
"LiftPegUpright-v1": {
|
| 51 |
+
"action_mse_to_best": 0.31969772514973566,
|
| 52 |
+
"expert_success_rate": 0.8865979381443299,
|
| 53 |
+
"num_groups": 97,
|
| 54 |
+
"oracle_success_rate": 0.9278350515463918,
|
| 55 |
+
"policy_expert_regret": 1.0792337135248578,
|
| 56 |
+
"policy_oracle_regret": 1.1056491912333006,
|
| 57 |
+
"policy_rollout_progress": 0.570563180084081,
|
| 58 |
+
"policy_rollout_success_rate": 0.2268041237113402,
|
| 59 |
+
"restore_max_error": 4.76837158203125e-07
|
| 60 |
+
},
|
| 61 |
+
"PickCube-v1": {
|
| 62 |
+
"action_mse_to_best": 0.3026706839625079,
|
| 63 |
+
"expert_success_rate": 0.9375,
|
| 64 |
+
"num_groups": 208,
|
| 65 |
+
"oracle_success_rate": 0.9471153846153846,
|
| 66 |
+
"policy_expert_regret": 0.9611832474788221,
|
| 67 |
+
"policy_oracle_regret": 0.9746061149053276,
|
| 68 |
+
"policy_rollout_progress": 0.6160833963479561,
|
| 69 |
+
"policy_rollout_success_rate": 0.34134615384615385,
|
| 70 |
+
"restore_max_error": 4.76837158203125e-07
|
| 71 |
+
},
|
| 72 |
+
"PullCube-v1": {
|
| 73 |
+
"action_mse_to_best": 0.6003992125972525,
|
| 74 |
+
"expert_success_rate": 0.19480519480519481,
|
| 75 |
+
"num_groups": 77,
|
| 76 |
+
"oracle_success_rate": 0.36363636363636365,
|
| 77 |
+
"policy_expert_regret": 0.19700447897209072,
|
| 78 |
+
"policy_oracle_regret": 0.4139321176849257,
|
| 79 |
+
"policy_rollout_progress": 0.32554851807109625,
|
| 80 |
+
"policy_rollout_success_rate": 0.22077922077922077,
|
| 81 |
+
"restore_max_error": 4.76837158203125e-07
|
| 82 |
+
},
|
| 83 |
+
"PushCube-v1": {
|
| 84 |
+
"action_mse_to_best": 0.3656082003912393,
|
| 85 |
+
"expert_success_rate": 0.8617021276595744,
|
| 86 |
+
"num_groups": 94,
|
| 87 |
+
"oracle_success_rate": 0.9893617021276596,
|
| 88 |
+
"policy_expert_regret": 0.5719942746961371,
|
| 89 |
+
"policy_oracle_regret": 0.66845020953011,
|
| 90 |
+
"policy_rollout_progress": 0.682613620257124,
|
| 91 |
+
"policy_rollout_success_rate": 0.648936170212766,
|
| 92 |
+
"restore_max_error": 4.76837158203125e-07
|
| 93 |
+
},
|
| 94 |
+
"StackCube-v1": {
|
| 95 |
+
"action_mse_to_best": 0.4427097219070702,
|
| 96 |
+
"expert_success_rate": 0.6767676767676768,
|
| 97 |
+
"num_groups": 99,
|
| 98 |
+
"oracle_success_rate": 0.8585858585858586,
|
| 99 |
+
"policy_expert_regret": 0.9762899511390262,
|
| 100 |
+
"policy_oracle_regret": 1.1504495456965282,
|
| 101 |
+
"policy_rollout_progress": 0.4381369689498285,
|
| 102 |
+
"policy_rollout_success_rate": 0.23232323232323232,
|
| 103 |
+
"restore_max_error": 3.948807716369629e-07
|
| 104 |
+
}
|
| 105 |
+
}
|
| 106 |
+
},
|
| 107 |
+
{
|
| 108 |
+
"seed": 1,
|
| 109 |
+
"path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/near_miss_policy_bc5/seed_1/policy_rollout_retrieval_residual_scale0p35_safe_types_margin0p30.json",
|
| 110 |
+
"num_groups": 575,
|
| 111 |
+
"selection_mode": "retrieval_residual",
|
| 112 |
+
"num_candidates": 16,
|
| 113 |
+
"candidate_sigma": 0.0,
|
| 114 |
+
"selection_margin": 0.3,
|
| 115 |
+
"field_optim_steps": 0,
|
| 116 |
+
"field_optim_step_size": 0.0,
|
| 117 |
+
"field_optim_trust_radius": 0.0,
|
| 118 |
+
"field_optim_l2_penalty": 0.0,
|
| 119 |
+
"retrieval_neighbors": 1,
|
| 120 |
+
"retrieval_metric": "raw",
|
| 121 |
+
"retrieval_type_min_success": 0.0,
|
| 122 |
+
"retrieval_residual_scale": 0.35,
|
| 123 |
+
"retrieval_residual_anchor": "expert",
|
| 124 |
+
"policy_rollout_success_rate": 0.3339130434782609,
|
| 125 |
+
"policy_rollout_progress": 0.5594285789812389,
|
| 126 |
+
"oracle_success_rate": 0.8695652173913043,
|
| 127 |
+
"action_mse_to_best": 0.3878339341703964,
|
| 128 |
+
"best_policy_val": {
|
| 129 |
+
"bc_loss": 0.1315989200439718,
|
| 130 |
+
"field_effect_loss": 0.010110370814800262,
|
| 131 |
+
"field_potential_loss": 0.31546536915832096,
|
| 132 |
+
"field_preference_loss": 0.4960637440284093,
|
| 133 |
+
"lattice_edges": 3833.3333333333335,
|
| 134 |
+
"progress_mae": 0.19023226698239645,
|
| 135 |
+
"rank_acc": 0.8255469501018524,
|
| 136 |
+
"rank_loss": 0.4960531195004781,
|
| 137 |
+
"regret_mae": 0.42156532737943864,
|
| 138 |
+
"success_accuracy": 0.8807473679383596,
|
| 139 |
+
"total_loss": 1.561984618504842
|
| 140 |
+
},
|
| 141 |
+
"per_task": {
|
| 142 |
+
"LiftPegUpright-v1": {
|
| 143 |
+
"action_mse_to_best": 0.32557108186482064,
|
| 144 |
+
"expert_success_rate": 0.8584070796460177,
|
| 145 |
+
"num_groups": 113,
|
| 146 |
+
"oracle_success_rate": 0.9380530973451328,
|
| 147 |
+
"policy_expert_regret": 0.987694250535121,
|
| 148 |
+
"policy_oracle_regret": 1.066896405631462,
|
| 149 |
+
"policy_rollout_progress": 0.6352153201546289,
|
| 150 |
+
"policy_rollout_success_rate": 0.23008849557522124,
|
| 151 |
+
"restore_max_error": 4.76837158203125e-07
|
| 152 |
+
},
|
| 153 |
+
"PickCube-v1": {
|
| 154 |
+
"action_mse_to_best": 0.3196464820926928,
|
| 155 |
+
"expert_success_rate": 0.9402173913043478,
|
| 156 |
+
"num_groups": 184,
|
| 157 |
+
"oracle_success_rate": 0.9456521739130435,
|
| 158 |
+
"policy_expert_regret": 1.0787854827029388,
|
| 159 |
+
"policy_oracle_regret": 1.0788094871778688,
|
| 160 |
+
"policy_rollout_progress": 0.5718511491305316,
|
| 161 |
+
"policy_rollout_success_rate": 0.28804347826086957,
|
| 162 |
+
"restore_max_error": 4.76837158203125e-07
|
| 163 |
+
},
|
| 164 |
+
"PullCube-v1": {
|
| 165 |
+
"action_mse_to_best": 0.57169899932648,
|
| 166 |
+
"expert_success_rate": 0.25,
|
| 167 |
+
"num_groups": 76,
|
| 168 |
+
"oracle_success_rate": 0.40789473684210525,
|
| 169 |
+
"policy_expert_regret": 0.3012117788587746,
|
| 170 |
+
"policy_oracle_regret": 0.5738564925758463,
|
| 171 |
+
"policy_rollout_progress": 0.2650559822979726,
|
| 172 |
+
"policy_rollout_success_rate": 0.15789473684210525,
|
| 173 |
+
"restore_max_error": 4.76837158203125e-07
|
| 174 |
+
},
|
| 175 |
+
"PushCube-v1": {
|
| 176 |
+
"action_mse_to_best": 0.3568917483024232,
|
| 177 |
+
"expert_success_rate": 0.8198198198198198,
|
| 178 |
+
"num_groups": 111,
|
| 179 |
+
"oracle_success_rate": 1.0,
|
| 180 |
+
"policy_expert_regret": 0.3483100165654947,
|
| 181 |
+
"policy_oracle_regret": 0.4142704948380187,
|
| 182 |
+
"policy_rollout_progress": 0.8019457213781975,
|
| 183 |
+
"policy_rollout_success_rate": 0.7837837837837838,
|
| 184 |
+
"restore_max_error": 4.76837158203125e-07
|
| 185 |
+
},
|
| 186 |
+
"StackCube-v1": {
|
| 187 |
+
"action_mse_to_best": 0.4872079030968822,
|
| 188 |
+
"expert_success_rate": 0.6923076923076923,
|
| 189 |
+
"num_groups": 91,
|
| 190 |
+
"oracle_success_rate": 0.8571428571428571,
|
| 191 |
+
"policy_expert_regret": 1.1282239404025969,
|
| 192 |
+
"policy_oracle_regret": 1.3003716945320696,
|
| 193 |
+
"policy_rollout_progress": 0.3902336326274243,
|
| 194 |
+
"policy_rollout_success_rate": 0.15384615384615385,
|
| 195 |
+
"restore_max_error": 3.948807716369629e-07
|
| 196 |
+
}
|
| 197 |
+
}
|
| 198 |
+
},
|
| 199 |
+
{
|
| 200 |
+
"seed": 2,
|
| 201 |
+
"path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/near_miss_policy_bc5/seed_2/policy_rollout_retrieval_residual_scale0p35_safe_types_margin0p30.json",
|
| 202 |
+
"num_groups": 575,
|
| 203 |
+
"selection_mode": "retrieval_residual",
|
| 204 |
+
"num_candidates": 16,
|
| 205 |
+
"candidate_sigma": 0.0,
|
| 206 |
+
"selection_margin": 0.3,
|
| 207 |
+
"field_optim_steps": 0,
|
| 208 |
+
"field_optim_step_size": 0.0,
|
| 209 |
+
"field_optim_trust_radius": 0.0,
|
| 210 |
+
"field_optim_l2_penalty": 0.0,
|
| 211 |
+
"retrieval_neighbors": 1,
|
| 212 |
+
"retrieval_metric": "raw",
|
| 213 |
+
"retrieval_type_min_success": 0.0,
|
| 214 |
+
"retrieval_residual_scale": 0.35,
|
| 215 |
+
"retrieval_residual_anchor": "expert",
|
| 216 |
+
"policy_rollout_success_rate": 0.36695652173913046,
|
| 217 |
+
"policy_rollout_progress": 0.5817452996245771,
|
| 218 |
+
"oracle_success_rate": 0.8765217391304347,
|
| 219 |
+
"action_mse_to_best": 0.4154942298947793,
|
| 220 |
+
"best_policy_val": {
|
| 221 |
+
"bc_loss": 0.11367896075050037,
|
| 222 |
+
"field_effect_loss": 0.009670218582161598,
|
| 223 |
+
"field_potential_loss": 0.2641640139950646,
|
| 224 |
+
"field_preference_loss": 0.5130490180518892,
|
| 225 |
+
"lattice_edges": 3833.3333333333335,
|
| 226 |
+
"progress_mae": 0.2021729110015763,
|
| 227 |
+
"rank_acc": 0.8333857821093665,
|
| 228 |
+
"rank_loss": 0.5130119257503085,
|
| 229 |
+
"regret_mae": 0.3958987047274907,
|
| 230 |
+
"success_accuracy": 0.8680730561415354,
|
| 231 |
+
"total_loss": 1.4394984311527677
|
| 232 |
+
},
|
| 233 |
+
"per_task": {
|
| 234 |
+
"LiftPegUpright-v1": {
|
| 235 |
+
"action_mse_to_best": 0.35361564260771655,
|
| 236 |
+
"expert_success_rate": 0.8229166666666666,
|
| 237 |
+
"num_groups": 96,
|
| 238 |
+
"oracle_success_rate": 0.9270833333333334,
|
| 239 |
+
"policy_expert_regret": 0.8470402403424183,
|
| 240 |
+
"policy_oracle_regret": 0.9533149556567272,
|
| 241 |
+
"policy_rollout_progress": 0.6469602342694998,
|
| 242 |
+
"policy_rollout_success_rate": 0.3333333333333333,
|
| 243 |
+
"restore_max_error": 3.5762786865234375e-07
|
| 244 |
+
},
|
| 245 |
+
"PickCube-v1": {
|
| 246 |
+
"action_mse_to_best": 0.3353338512695498,
|
| 247 |
+
"expert_success_rate": 0.9444444444444444,
|
| 248 |
+
"num_groups": 198,
|
| 249 |
+
"oracle_success_rate": 0.9595959595959596,
|
| 250 |
+
"policy_expert_regret": 0.995343850586902,
|
| 251 |
+
"policy_oracle_regret": 1.0024273011156104,
|
| 252 |
+
"policy_rollout_progress": 0.6312458732475837,
|
| 253 |
+
"policy_rollout_success_rate": 0.3282828282828283,
|
| 254 |
+
"restore_max_error": 4.76837158203125e-07
|
| 255 |
+
},
|
| 256 |
+
"PullCube-v1": {
|
| 257 |
+
"action_mse_to_best": 0.6295425814059046,
|
| 258 |
+
"expert_success_rate": 0.24444444444444444,
|
| 259 |
+
"num_groups": 90,
|
| 260 |
+
"oracle_success_rate": 0.4666666666666667,
|
| 261 |
+
"policy_expert_regret": 0.2935961757958972,
|
| 262 |
+
"policy_oracle_regret": 0.5007589669760111,
|
| 263 |
+
"policy_rollout_progress": 0.3386433102121575,
|
| 264 |
+
"policy_rollout_success_rate": 0.23333333333333334,
|
| 265 |
+
"restore_max_error": 4.0978193283081055e-07
|
| 266 |
+
},
|
| 267 |
+
"PushCube-v1": {
|
| 268 |
+
"action_mse_to_best": 0.3798191750256142,
|
| 269 |
+
"expert_success_rate": 0.8514851485148515,
|
| 270 |
+
"num_groups": 101,
|
| 271 |
+
"oracle_success_rate": 1.0,
|
| 272 |
+
"policy_expert_regret": 0.36240387021905124,
|
| 273 |
+
"policy_oracle_regret": 0.43505349238910296,
|
| 274 |
+
"policy_rollout_progress": 0.7926692798881247,
|
| 275 |
+
"policy_rollout_success_rate": 0.7722772277227723,
|
| 276 |
+
"restore_max_error": 4.76837158203125e-07
|
| 277 |
+
},
|
| 278 |
+
"StackCube-v1": {
|
| 279 |
+
"action_mse_to_best": 0.48383787715186677,
|
| 280 |
+
"expert_success_rate": 0.7666666666666667,
|
| 281 |
+
"num_groups": 90,
|
| 282 |
+
"oracle_success_rate": 0.9111111111111111,
|
| 283 |
+
"policy_expert_regret": 1.15659624768628,
|
| 284 |
+
"policy_oracle_regret": 1.3056684128112264,
|
| 285 |
+
"policy_rollout_progress": 0.40967985226048365,
|
| 286 |
+
"policy_rollout_success_rate": 0.16666666666666666,
|
| 287 |
+
"restore_max_error": 4.76837158203125e-07
|
| 288 |
+
}
|
| 289 |
+
}
|
| 290 |
+
}
|
| 291 |
+
]
|
| 292 |
+
}
|
results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_scale0p35_safe_types_margin0p30_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_scale0p35_safe_types_margin0p30.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: 34.61% +/- 1.82%
|
| 11 |
+
Gain vs h=16 rank checkpoint: +4.87%
|
| 12 |
+
Mean progress: 56.36%
|
| 13 |
+
Mean action MSE to best: 0.394
|
| 14 |
+
|
| 15 |
+
| seed | mode | k | retrieval K | retrieval metric | residual anchor | min type success | residual scale | margin | sigma | opt steps | trust | success | progress | oracle | action MSE |
|
| 16 |
+
|---:|---|---:|---:|---|---|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|
|
| 17 |
+
| 0 | retrieval_residual | 16 | 1 | raw | expert | 0.00 | 0.35 | 0.300 | 0.00 | 0 | 0.00 | 33.74% | 54.97% | 85.74% | 0.380 |
|
| 18 |
+
| 1 | retrieval_residual | 16 | 1 | raw | expert | 0.00 | 0.35 | 0.300 | 0.00 | 0 | 0.00 | 33.39% | 55.94% | 86.96% | 0.388 |
|
| 19 |
+
| 2 | retrieval_residual | 16 | 1 | raw | expert | 0.00 | 0.35 | 0.300 | 0.00 | 0 | 0.00 | 36.70% | 58.17% | 87.65% | 0.415 |
|
results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_scale0p35_safe_types_margin0p40_summary.json
ADDED
|
@@ -0,0 +1,292 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 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_scale0p35_safe_types_margin0p40.json",
|
| 5 |
+
"num_completed": 3,
|
| 6 |
+
"baseline_h4_policy_success": 0.2967,
|
| 7 |
+
"baseline_h16_rank_checkpoint_success": 0.29739130434782607,
|
| 8 |
+
"mean_success": 0.34492753623188405,
|
| 9 |
+
"std_success": 0.019156767120995153,
|
| 10 |
+
"mean_progress": 0.562753911826103,
|
| 11 |
+
"mean_action_mse_to_best": 0.3946052004461703,
|
| 12 |
+
"gain_vs_h4": 0.04822753623188403,
|
| 13 |
+
"gain_vs_h16_rank_checkpoint": 0.04753623188405798,
|
| 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_scale0p35_safe_types_margin0p40.json",
|
| 18 |
+
"num_groups": 575,
|
| 19 |
+
"selection_mode": "retrieval_residual",
|
| 20 |
+
"num_candidates": 16,
|
| 21 |
+
"candidate_sigma": 0.0,
|
| 22 |
+
"selection_margin": 0.4,
|
| 23 |
+
"field_optim_steps": 0,
|
| 24 |
+
"field_optim_step_size": 0.0,
|
| 25 |
+
"field_optim_trust_radius": 0.0,
|
| 26 |
+
"field_optim_l2_penalty": 0.0,
|
| 27 |
+
"retrieval_neighbors": 1,
|
| 28 |
+
"retrieval_metric": "raw",
|
| 29 |
+
"retrieval_type_min_success": 0.0,
|
| 30 |
+
"retrieval_residual_scale": 0.35,
|
| 31 |
+
"retrieval_residual_anchor": "expert",
|
| 32 |
+
"policy_rollout_success_rate": 0.3356521739130435,
|
| 33 |
+
"policy_rollout_progress": 0.5482615261699032,
|
| 34 |
+
"oracle_success_rate": 0.8573913043478261,
|
| 35 |
+
"action_mse_to_best": 0.38090274694092247,
|
| 36 |
+
"best_policy_val": {
|
| 37 |
+
"bc_loss": 0.13721593966086706,
|
| 38 |
+
"field_effect_loss": 0.009290305380192068,
|
| 39 |
+
"field_potential_loss": 0.2666468388504452,
|
| 40 |
+
"field_preference_loss": 0.5130573478009965,
|
| 41 |
+
"lattice_edges": 3833.3333333333335,
|
| 42 |
+
"progress_mae": 0.1933159919248687,
|
| 43 |
+
"rank_acc": 0.8265031774838766,
|
| 44 |
+
"rank_loss": 0.5130523675017886,
|
| 45 |
+
"regret_mae": 0.3756548762321472,
|
| 46 |
+
"success_accuracy": 0.8773836526605818,
|
| 47 |
+
"total_loss": 1.5581054819954767
|
| 48 |
+
},
|
| 49 |
+
"per_task": {
|
| 50 |
+
"LiftPegUpright-v1": {
|
| 51 |
+
"action_mse_to_best": 0.31969772514973566,
|
| 52 |
+
"expert_success_rate": 0.8865979381443299,
|
| 53 |
+
"num_groups": 97,
|
| 54 |
+
"oracle_success_rate": 0.9278350515463918,
|
| 55 |
+
"policy_expert_regret": 1.0792337135248578,
|
| 56 |
+
"policy_oracle_regret": 1.1056491912333006,
|
| 57 |
+
"policy_rollout_progress": 0.570563180084081,
|
| 58 |
+
"policy_rollout_success_rate": 0.2268041237113402,
|
| 59 |
+
"restore_max_error": 4.76837158203125e-07
|
| 60 |
+
},
|
| 61 |
+
"PickCube-v1": {
|
| 62 |
+
"action_mse_to_best": 0.3026178333406838,
|
| 63 |
+
"expert_success_rate": 0.9375,
|
| 64 |
+
"num_groups": 208,
|
| 65 |
+
"oracle_success_rate": 0.9471153846153846,
|
| 66 |
+
"policy_expert_regret": 0.9612465917419356,
|
| 67 |
+
"policy_oracle_regret": 0.974669459168441,
|
| 68 |
+
"policy_rollout_progress": 0.6160200520848426,
|
| 69 |
+
"policy_rollout_success_rate": 0.34134615384615385,
|
| 70 |
+
"restore_max_error": 4.76837158203125e-07
|
| 71 |
+
},
|
| 72 |
+
"PullCube-v1": {
|
| 73 |
+
"action_mse_to_best": 0.6086812663774985,
|
| 74 |
+
"expert_success_rate": 0.19480519480519481,
|
| 75 |
+
"num_groups": 77,
|
| 76 |
+
"oracle_success_rate": 0.36363636363636365,
|
| 77 |
+
"policy_expert_regret": 0.22065381407314402,
|
| 78 |
+
"policy_oracle_regret": 0.43776196022759883,
|
| 79 |
+
"policy_rollout_progress": 0.3147056885154361,
|
| 80 |
+
"policy_rollout_success_rate": 0.2077922077922078,
|
| 81 |
+
"restore_max_error": 4.76837158203125e-07
|
| 82 |
+
},
|
| 83 |
+
"PushCube-v1": {
|
| 84 |
+
"action_mse_to_best": 0.3656082003912393,
|
| 85 |
+
"expert_success_rate": 0.8617021276595744,
|
| 86 |
+
"num_groups": 94,
|
| 87 |
+
"oracle_success_rate": 0.9893617021276596,
|
| 88 |
+
"policy_expert_regret": 0.5719942746961371,
|
| 89 |
+
"policy_oracle_regret": 0.66845020953011,
|
| 90 |
+
"policy_rollout_progress": 0.682613620257124,
|
| 91 |
+
"policy_rollout_success_rate": 0.648936170212766,
|
| 92 |
+
"restore_max_error": 4.76837158203125e-07
|
| 93 |
+
},
|
| 94 |
+
"StackCube-v1": {
|
| 95 |
+
"action_mse_to_best": 0.4427097219070702,
|
| 96 |
+
"expert_success_rate": 0.6767676767676768,
|
| 97 |
+
"num_groups": 99,
|
| 98 |
+
"oracle_success_rate": 0.8585858585858586,
|
| 99 |
+
"policy_expert_regret": 0.9762899511390262,
|
| 100 |
+
"policy_oracle_regret": 1.1504495456965282,
|
| 101 |
+
"policy_rollout_progress": 0.4381369689498285,
|
| 102 |
+
"policy_rollout_success_rate": 0.23232323232323232,
|
| 103 |
+
"restore_max_error": 3.948807716369629e-07
|
| 104 |
+
}
|
| 105 |
+
}
|
| 106 |
+
},
|
| 107 |
+
{
|
| 108 |
+
"seed": 1,
|
| 109 |
+
"path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/near_miss_policy_bc5/seed_1/policy_rollout_retrieval_residual_scale0p35_safe_types_margin0p40.json",
|
| 110 |
+
"num_groups": 575,
|
| 111 |
+
"selection_mode": "retrieval_residual",
|
| 112 |
+
"num_candidates": 16,
|
| 113 |
+
"candidate_sigma": 0.0,
|
| 114 |
+
"selection_margin": 0.4,
|
| 115 |
+
"field_optim_steps": 0,
|
| 116 |
+
"field_optim_step_size": 0.0,
|
| 117 |
+
"field_optim_trust_radius": 0.0,
|
| 118 |
+
"field_optim_l2_penalty": 0.0,
|
| 119 |
+
"retrieval_neighbors": 1,
|
| 120 |
+
"retrieval_metric": "raw",
|
| 121 |
+
"retrieval_type_min_success": 0.0,
|
| 122 |
+
"retrieval_residual_scale": 0.35,
|
| 123 |
+
"retrieval_residual_anchor": "expert",
|
| 124 |
+
"policy_rollout_success_rate": 0.33217391304347826,
|
| 125 |
+
"policy_rollout_progress": 0.5581722561043242,
|
| 126 |
+
"oracle_success_rate": 0.8695652173913043,
|
| 127 |
+
"action_mse_to_best": 0.3881500020526026,
|
| 128 |
+
"best_policy_val": {
|
| 129 |
+
"bc_loss": 0.1315989200439718,
|
| 130 |
+
"field_effect_loss": 0.010110370814800262,
|
| 131 |
+
"field_potential_loss": 0.31546536915832096,
|
| 132 |
+
"field_preference_loss": 0.4960637440284093,
|
| 133 |
+
"lattice_edges": 3833.3333333333335,
|
| 134 |
+
"progress_mae": 0.19023226698239645,
|
| 135 |
+
"rank_acc": 0.8255469501018524,
|
| 136 |
+
"rank_loss": 0.4960531195004781,
|
| 137 |
+
"regret_mae": 0.42156532737943864,
|
| 138 |
+
"success_accuracy": 0.8807473679383596,
|
| 139 |
+
"total_loss": 1.561984618504842
|
| 140 |
+
},
|
| 141 |
+
"per_task": {
|
| 142 |
+
"LiftPegUpright-v1": {
|
| 143 |
+
"action_mse_to_best": 0.32557108186482064,
|
| 144 |
+
"expert_success_rate": 0.8584070796460177,
|
| 145 |
+
"num_groups": 113,
|
| 146 |
+
"oracle_success_rate": 0.9380530973451328,
|
| 147 |
+
"policy_expert_regret": 0.987694250535121,
|
| 148 |
+
"policy_oracle_regret": 1.066896405631462,
|
| 149 |
+
"policy_rollout_progress": 0.6352153201546289,
|
| 150 |
+
"policy_rollout_success_rate": 0.23008849557522124,
|
| 151 |
+
"restore_max_error": 4.76837158203125e-07
|
| 152 |
+
},
|
| 153 |
+
"PickCube-v1": {
|
| 154 |
+
"action_mse_to_best": 0.3195808531675974,
|
| 155 |
+
"expert_success_rate": 0.9402173913043478,
|
| 156 |
+
"num_groups": 184,
|
| 157 |
+
"oracle_success_rate": 0.9456521739130435,
|
| 158 |
+
"policy_expert_regret": 1.0786933760340933,
|
| 159 |
+
"policy_oracle_regret": 1.0787173805090233,
|
| 160 |
+
"policy_rollout_progress": 0.5719432557993771,
|
| 161 |
+
"policy_rollout_success_rate": 0.28804347826086957,
|
| 162 |
+
"restore_max_error": 4.76837158203125e-07
|
| 163 |
+
},
|
| 164 |
+
"PullCube-v1": {
|
| 165 |
+
"action_mse_to_best": 0.575498066842556,
|
| 166 |
+
"expert_success_rate": 0.25,
|
| 167 |
+
"num_groups": 76,
|
| 168 |
+
"oracle_success_rate": 0.40789473684210525,
|
| 169 |
+
"policy_expert_regret": 0.3254193017798427,
|
| 170 |
+
"policy_oracle_regret": 0.5966785812113238,
|
| 171 |
+
"policy_rollout_progress": 0.2553917883993371,
|
| 172 |
+
"policy_rollout_success_rate": 0.14473684210526316,
|
| 173 |
+
"restore_max_error": 4.76837158203125e-07
|
| 174 |
+
},
|
| 175 |
+
"PushCube-v1": {
|
| 176 |
+
"action_mse_to_best": 0.3568917483024232,
|
| 177 |
+
"expert_success_rate": 0.8198198198198198,
|
| 178 |
+
"num_groups": 111,
|
| 179 |
+
"oracle_success_rate": 1.0,
|
| 180 |
+
"policy_expert_regret": 0.3483100165654947,
|
| 181 |
+
"policy_oracle_regret": 0.4142704948380187,
|
| 182 |
+
"policy_rollout_progress": 0.8019457213781975,
|
| 183 |
+
"policy_rollout_success_rate": 0.7837837837837838,
|
| 184 |
+
"restore_max_error": 4.76837158203125e-07
|
| 185 |
+
},
|
| 186 |
+
"StackCube-v1": {
|
| 187 |
+
"action_mse_to_best": 0.4861648879679186,
|
| 188 |
+
"expert_success_rate": 0.6923076923076923,
|
| 189 |
+
"num_groups": 91,
|
| 190 |
+
"oracle_success_rate": 0.8571428571428571,
|
| 191 |
+
"policy_expert_regret": 1.128277287050918,
|
| 192 |
+
"policy_oracle_regret": 1.3004250411803906,
|
| 193 |
+
"policy_rollout_progress": 0.39018028597910326,
|
| 194 |
+
"policy_rollout_success_rate": 0.15384615384615385,
|
| 195 |
+
"restore_max_error": 3.948807716369629e-07
|
| 196 |
+
}
|
| 197 |
+
}
|
| 198 |
+
},
|
| 199 |
+
{
|
| 200 |
+
"seed": 2,
|
| 201 |
+
"path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/near_miss_policy_bc5/seed_2/policy_rollout_retrieval_residual_scale0p35_safe_types_margin0p40.json",
|
| 202 |
+
"num_groups": 575,
|
| 203 |
+
"selection_mode": "retrieval_residual",
|
| 204 |
+
"num_candidates": 16,
|
| 205 |
+
"candidate_sigma": 0.0,
|
| 206 |
+
"selection_margin": 0.4,
|
| 207 |
+
"field_optim_steps": 0,
|
| 208 |
+
"field_optim_step_size": 0.0,
|
| 209 |
+
"field_optim_trust_radius": 0.0,
|
| 210 |
+
"field_optim_l2_penalty": 0.0,
|
| 211 |
+
"retrieval_neighbors": 1,
|
| 212 |
+
"retrieval_metric": "raw",
|
| 213 |
+
"retrieval_type_min_success": 0.0,
|
| 214 |
+
"retrieval_residual_scale": 0.35,
|
| 215 |
+
"retrieval_residual_anchor": "expert",
|
| 216 |
+
"policy_rollout_success_rate": 0.36695652173913046,
|
| 217 |
+
"policy_rollout_progress": 0.5818279532040818,
|
| 218 |
+
"oracle_success_rate": 0.8765217391304347,
|
| 219 |
+
"action_mse_to_best": 0.4147628523449859,
|
| 220 |
+
"best_policy_val": {
|
| 221 |
+
"bc_loss": 0.11367896075050037,
|
| 222 |
+
"field_effect_loss": 0.009670218582161598,
|
| 223 |
+
"field_potential_loss": 0.2641640139950646,
|
| 224 |
+
"field_preference_loss": 0.5130490180518892,
|
| 225 |
+
"lattice_edges": 3833.3333333333335,
|
| 226 |
+
"progress_mae": 0.2021729110015763,
|
| 227 |
+
"rank_acc": 0.8333857821093665,
|
| 228 |
+
"rank_loss": 0.5130119257503085,
|
| 229 |
+
"regret_mae": 0.3958987047274907,
|
| 230 |
+
"success_accuracy": 0.8680730561415354,
|
| 231 |
+
"total_loss": 1.4394984311527677
|
| 232 |
+
},
|
| 233 |
+
"per_task": {
|
| 234 |
+
"LiftPegUpright-v1": {
|
| 235 |
+
"action_mse_to_best": 0.35361564260771655,
|
| 236 |
+
"expert_success_rate": 0.8229166666666666,
|
| 237 |
+
"num_groups": 96,
|
| 238 |
+
"oracle_success_rate": 0.9270833333333334,
|
| 239 |
+
"policy_expert_regret": 0.8470402403424183,
|
| 240 |
+
"policy_oracle_regret": 0.9533149556567272,
|
| 241 |
+
"policy_rollout_progress": 0.6469602342694998,
|
| 242 |
+
"policy_rollout_success_rate": 0.3333333333333333,
|
| 243 |
+
"restore_max_error": 3.5762786865234375e-07
|
| 244 |
+
},
|
| 245 |
+
"PickCube-v1": {
|
| 246 |
+
"action_mse_to_best": 0.33526025458493014,
|
| 247 |
+
"expert_success_rate": 0.9444444444444444,
|
| 248 |
+
"num_groups": 198,
|
| 249 |
+
"oracle_success_rate": 0.9595959595959596,
|
| 250 |
+
"policy_expert_regret": 0.995225173825718,
|
| 251 |
+
"policy_oracle_regret": 1.0023086243544264,
|
| 252 |
+
"policy_rollout_progress": 0.6313645500087678,
|
| 253 |
+
"policy_rollout_success_rate": 0.3282828282828283,
|
| 254 |
+
"restore_max_error": 4.76837158203125e-07
|
| 255 |
+
},
|
| 256 |
+
"PullCube-v1": {
|
| 257 |
+
"action_mse_to_best": 0.6250318042106099,
|
| 258 |
+
"expert_success_rate": 0.24444444444444444,
|
| 259 |
+
"num_groups": 90,
|
| 260 |
+
"oracle_success_rate": 0.4666666666666667,
|
| 261 |
+
"policy_expert_regret": 0.292973979962638,
|
| 262 |
+
"policy_oracle_regret": 0.5004919913148923,
|
| 263 |
+
"policy_rollout_progress": 0.3389102858732763,
|
| 264 |
+
"policy_rollout_success_rate": 0.23333333333333334,
|
| 265 |
+
"restore_max_error": 4.0978193283081055e-07
|
| 266 |
+
},
|
| 267 |
+
"PushCube-v1": {
|
| 268 |
+
"action_mse_to_best": 0.3798191750256142,
|
| 269 |
+
"expert_success_rate": 0.8514851485148515,
|
| 270 |
+
"num_groups": 101,
|
| 271 |
+
"oracle_success_rate": 1.0,
|
| 272 |
+
"policy_expert_regret": 0.36240387021905124,
|
| 273 |
+
"policy_oracle_regret": 0.43505349238910296,
|
| 274 |
+
"policy_rollout_progress": 0.7926692798881247,
|
| 275 |
+
"policy_rollout_success_rate": 0.7722772277227723,
|
| 276 |
+
"restore_max_error": 4.76837158203125e-07
|
| 277 |
+
},
|
| 278 |
+
"StackCube-v1": {
|
| 279 |
+
"action_mse_to_best": 0.48383787715186677,
|
| 280 |
+
"expert_success_rate": 0.7666666666666667,
|
| 281 |
+
"num_groups": 90,
|
| 282 |
+
"oracle_success_rate": 0.9111111111111111,
|
| 283 |
+
"policy_expert_regret": 1.15659624768628,
|
| 284 |
+
"policy_oracle_regret": 1.3056684128112264,
|
| 285 |
+
"policy_rollout_progress": 0.40967985226048365,
|
| 286 |
+
"policy_rollout_success_rate": 0.16666666666666666,
|
| 287 |
+
"restore_max_error": 4.76837158203125e-07
|
| 288 |
+
}
|
| 289 |
+
}
|
| 290 |
+
}
|
| 291 |
+
]
|
| 292 |
+
}
|
results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_scale0p35_safe_types_margin0p40_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_scale0p35_safe_types_margin0p40.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: 34.49% +/- 1.92%
|
| 11 |
+
Gain vs h=16 rank checkpoint: +4.75%
|
| 12 |
+
Mean progress: 56.28%
|
| 13 |
+
Mean action MSE to best: 0.395
|
| 14 |
+
|
| 15 |
+
| seed | mode | k | retrieval K | retrieval metric | residual anchor | min type success | residual scale | margin | sigma | opt steps | trust | success | progress | oracle | action MSE |
|
| 16 |
+
|---:|---|---:|---:|---|---|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|
|
| 17 |
+
| 0 | retrieval_residual | 16 | 1 | raw | expert | 0.00 | 0.35 | 0.400 | 0.00 | 0 | 0.00 | 33.57% | 54.83% | 85.74% | 0.381 |
|
| 18 |
+
| 1 | retrieval_residual | 16 | 1 | raw | expert | 0.00 | 0.35 | 0.400 | 0.00 | 0 | 0.00 | 33.22% | 55.82% | 86.96% | 0.388 |
|
| 19 |
+
| 2 | retrieval_residual | 16 | 1 | raw | expert | 0.00 | 0.35 | 0.400 | 0.00 | 0 | 0.00 | 36.70% | 58.18% | 87.65% | 0.415 |
|
results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_scale0p35_safe_types_margin0p60_summary.json
ADDED
|
@@ -0,0 +1,292 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 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_scale0p35_safe_types_margin0p60.json",
|
| 5 |
+
"num_completed": 3,
|
| 6 |
+
"baseline_h4_policy_success": 0.2967,
|
| 7 |
+
"baseline_h16_rank_checkpoint_success": 0.29739130434782607,
|
| 8 |
+
"mean_success": 0.34144927536231884,
|
| 9 |
+
"std_success": 0.01915676712099512,
|
| 10 |
+
"mean_progress": 0.5607530164459931,
|
| 11 |
+
"mean_action_mse_to_best": 0.39419831574827,
|
| 12 |
+
"gain_vs_h4": 0.04474927536231882,
|
| 13 |
+
"gain_vs_h16_rank_checkpoint": 0.04405797101449277,
|
| 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_scale0p35_safe_types_margin0p60.json",
|
| 18 |
+
"num_groups": 575,
|
| 19 |
+
"selection_mode": "retrieval_residual",
|
| 20 |
+
"num_candidates": 16,
|
| 21 |
+
"candidate_sigma": 0.0,
|
| 22 |
+
"selection_margin": 0.6,
|
| 23 |
+
"field_optim_steps": 0,
|
| 24 |
+
"field_optim_step_size": 0.0,
|
| 25 |
+
"field_optim_trust_radius": 0.0,
|
| 26 |
+
"field_optim_l2_penalty": 0.0,
|
| 27 |
+
"retrieval_neighbors": 1,
|
| 28 |
+
"retrieval_metric": "raw",
|
| 29 |
+
"retrieval_type_min_success": 0.0,
|
| 30 |
+
"retrieval_residual_scale": 0.35,
|
| 31 |
+
"retrieval_residual_anchor": "expert",
|
| 32 |
+
"policy_rollout_success_rate": 0.33217391304347826,
|
| 33 |
+
"policy_rollout_progress": 0.5467438860768047,
|
| 34 |
+
"oracle_success_rate": 0.8573913043478261,
|
| 35 |
+
"action_mse_to_best": 0.3807186156762359,
|
| 36 |
+
"best_policy_val": {
|
| 37 |
+
"bc_loss": 0.13721593966086706,
|
| 38 |
+
"field_effect_loss": 0.009290305380192068,
|
| 39 |
+
"field_potential_loss": 0.2666468388504452,
|
| 40 |
+
"field_preference_loss": 0.5130573478009965,
|
| 41 |
+
"lattice_edges": 3833.3333333333335,
|
| 42 |
+
"progress_mae": 0.1933159919248687,
|
| 43 |
+
"rank_acc": 0.8265031774838766,
|
| 44 |
+
"rank_loss": 0.5130523675017886,
|
| 45 |
+
"regret_mae": 0.3756548762321472,
|
| 46 |
+
"success_accuracy": 0.8773836526605818,
|
| 47 |
+
"total_loss": 1.5581054819954767
|
| 48 |
+
},
|
| 49 |
+
"per_task": {
|
| 50 |
+
"LiftPegUpright-v1": {
|
| 51 |
+
"action_mse_to_best": 0.31969772514973566,
|
| 52 |
+
"expert_success_rate": 0.8865979381443299,
|
| 53 |
+
"num_groups": 97,
|
| 54 |
+
"oracle_success_rate": 0.9278350515463918,
|
| 55 |
+
"policy_expert_regret": 1.0792337135248578,
|
| 56 |
+
"policy_oracle_regret": 1.1056491912333006,
|
| 57 |
+
"policy_rollout_progress": 0.570563180084081,
|
| 58 |
+
"policy_rollout_success_rate": 0.2268041237113402,
|
| 59 |
+
"restore_max_error": 4.76837158203125e-07
|
| 60 |
+
},
|
| 61 |
+
"PickCube-v1": {
|
| 62 |
+
"action_mse_to_best": 0.30251795243328583,
|
| 63 |
+
"expert_success_rate": 0.9375,
|
| 64 |
+
"num_groups": 208,
|
| 65 |
+
"oracle_success_rate": 0.9471153846153846,
|
| 66 |
+
"policy_expert_regret": 0.9750573756531454,
|
| 67 |
+
"policy_oracle_regret": 0.9884802430796509,
|
| 68 |
+
"policy_rollout_progress": 0.6118246527890173,
|
| 69 |
+
"policy_rollout_success_rate": 0.3317307692307692,
|
| 70 |
+
"restore_max_error": 4.76837158203125e-07
|
| 71 |
+
},
|
| 72 |
+
"PullCube-v1": {
|
| 73 |
+
"action_mse_to_best": 0.6086052433623896,
|
| 74 |
+
"expert_success_rate": 0.19480519480519481,
|
| 75 |
+
"num_groups": 77,
|
| 76 |
+
"oracle_success_rate": 0.36363636363636365,
|
| 77 |
+
"policy_expert_regret": 0.22065381407314402,
|
| 78 |
+
"policy_oracle_regret": 0.43776196022759883,
|
| 79 |
+
"policy_rollout_progress": 0.3147056885154361,
|
| 80 |
+
"policy_rollout_success_rate": 0.2077922077922078,
|
| 81 |
+
"restore_max_error": 4.76837158203125e-07
|
| 82 |
+
},
|
| 83 |
+
"PushCube-v1": {
|
| 84 |
+
"action_mse_to_best": 0.36476515277110516,
|
| 85 |
+
"expert_success_rate": 0.8617021276595744,
|
| 86 |
+
"num_groups": 94,
|
| 87 |
+
"oracle_success_rate": 0.9893617021276596,
|
| 88 |
+
"policy_expert_regret": 0.5719942746961371,
|
| 89 |
+
"policy_oracle_regret": 0.66845020953011,
|
| 90 |
+
"policy_rollout_progress": 0.682613620257124,
|
| 91 |
+
"policy_rollout_success_rate": 0.648936170212766,
|
| 92 |
+
"restore_max_error": 4.76837158203125e-07
|
| 93 |
+
},
|
| 94 |
+
"StackCube-v1": {
|
| 95 |
+
"action_mse_to_best": 0.4427097219070702,
|
| 96 |
+
"expert_success_rate": 0.6767676767676768,
|
| 97 |
+
"num_groups": 99,
|
| 98 |
+
"oracle_success_rate": 0.8585858585858586,
|
| 99 |
+
"policy_expert_regret": 0.9762899511390262,
|
| 100 |
+
"policy_oracle_regret": 1.1504495456965282,
|
| 101 |
+
"policy_rollout_progress": 0.4381369689498285,
|
| 102 |
+
"policy_rollout_success_rate": 0.23232323232323232,
|
| 103 |
+
"restore_max_error": 3.948807716369629e-07
|
| 104 |
+
}
|
| 105 |
+
}
|
| 106 |
+
},
|
| 107 |
+
{
|
| 108 |
+
"seed": 1,
|
| 109 |
+
"path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/near_miss_policy_bc5/seed_1/policy_rollout_retrieval_residual_scale0p35_safe_types_margin0p60.json",
|
| 110 |
+
"num_groups": 575,
|
| 111 |
+
"selection_mode": "retrieval_residual",
|
| 112 |
+
"num_candidates": 16,
|
| 113 |
+
"candidate_sigma": 0.0,
|
| 114 |
+
"selection_margin": 0.6,
|
| 115 |
+
"field_optim_steps": 0,
|
| 116 |
+
"field_optim_step_size": 0.0,
|
| 117 |
+
"field_optim_trust_radius": 0.0,
|
| 118 |
+
"field_optim_l2_penalty": 0.0,
|
| 119 |
+
"retrieval_neighbors": 1,
|
| 120 |
+
"retrieval_metric": "raw",
|
| 121 |
+
"retrieval_type_min_success": 0.0,
|
| 122 |
+
"retrieval_residual_scale": 0.35,
|
| 123 |
+
"retrieval_residual_anchor": "expert",
|
| 124 |
+
"policy_rollout_success_rate": 0.32869565217391306,
|
| 125 |
+
"policy_rollout_progress": 0.5560782212537267,
|
| 126 |
+
"oracle_success_rate": 0.8695652173913043,
|
| 127 |
+
"action_mse_to_best": 0.3876221215757339,
|
| 128 |
+
"best_policy_val": {
|
| 129 |
+
"bc_loss": 0.1315989200439718,
|
| 130 |
+
"field_effect_loss": 0.010110370814800262,
|
| 131 |
+
"field_potential_loss": 0.31546536915832096,
|
| 132 |
+
"field_preference_loss": 0.4960637440284093,
|
| 133 |
+
"lattice_edges": 3833.3333333333335,
|
| 134 |
+
"progress_mae": 0.19023226698239645,
|
| 135 |
+
"rank_acc": 0.8255469501018524,
|
| 136 |
+
"rank_loss": 0.4960531195004781,
|
| 137 |
+
"regret_mae": 0.42156532737943864,
|
| 138 |
+
"success_accuracy": 0.8807473679383596,
|
| 139 |
+
"total_loss": 1.561984618504842
|
| 140 |
+
},
|
| 141 |
+
"per_task": {
|
| 142 |
+
"LiftPegUpright-v1": {
|
| 143 |
+
"action_mse_to_best": 0.32557108186482064,
|
| 144 |
+
"expert_success_rate": 0.8584070796460177,
|
| 145 |
+
"num_groups": 113,
|
| 146 |
+
"oracle_success_rate": 0.9380530973451328,
|
| 147 |
+
"policy_expert_regret": 0.987694250535121,
|
| 148 |
+
"policy_oracle_regret": 1.066896405631462,
|
| 149 |
+
"policy_rollout_progress": 0.6352153201546289,
|
| 150 |
+
"policy_rollout_success_rate": 0.23008849557522124,
|
| 151 |
+
"restore_max_error": 4.76837158203125e-07
|
| 152 |
+
},
|
| 153 |
+
"PickCube-v1": {
|
| 154 |
+
"action_mse_to_best": 0.319557456076955,
|
| 155 |
+
"expert_success_rate": 0.9402173913043478,
|
| 156 |
+
"num_groups": 184,
|
| 157 |
+
"oracle_success_rate": 0.9456521739130435,
|
| 158 |
+
"policy_expert_regret": 1.0864933169467133,
|
| 159 |
+
"policy_oracle_regret": 1.0865173214216433,
|
| 160 |
+
"policy_rollout_progress": 0.5695780974954529,
|
| 161 |
+
"policy_rollout_success_rate": 0.2826086956521739,
|
| 162 |
+
"restore_max_error": 4.76837158203125e-07
|
| 163 |
+
},
|
| 164 |
+
"PullCube-v1": {
|
| 165 |
+
"action_mse_to_best": 0.5715608798751706,
|
| 166 |
+
"expert_success_rate": 0.25,
|
| 167 |
+
"num_groups": 76,
|
| 168 |
+
"oracle_success_rate": 0.40789473684210525,
|
| 169 |
+
"policy_expert_regret": 0.3506602184405844,
|
| 170 |
+
"policy_oracle_regret": 0.619953330042527,
|
| 171 |
+
"policy_rollout_progress": 0.24527493430497616,
|
| 172 |
+
"policy_rollout_success_rate": 0.13157894736842105,
|
| 173 |
+
"restore_max_error": 4.76837158203125e-07
|
| 174 |
+
},
|
| 175 |
+
"PushCube-v1": {
|
| 176 |
+
"action_mse_to_best": 0.3568917483024232,
|
| 177 |
+
"expert_success_rate": 0.8198198198198198,
|
| 178 |
+
"num_groups": 111,
|
| 179 |
+
"oracle_success_rate": 1.0,
|
| 180 |
+
"policy_expert_regret": 0.3483100165654947,
|
| 181 |
+
"policy_oracle_regret": 0.4142704948380187,
|
| 182 |
+
"policy_rollout_progress": 0.8019457213781975,
|
| 183 |
+
"policy_rollout_success_rate": 0.7837837837837838,
|
| 184 |
+
"restore_max_error": 4.76837158203125e-07
|
| 185 |
+
},
|
| 186 |
+
"StackCube-v1": {
|
| 187 |
+
"action_mse_to_best": 0.4861648879679186,
|
| 188 |
+
"expert_success_rate": 0.6923076923076923,
|
| 189 |
+
"num_groups": 91,
|
| 190 |
+
"oracle_success_rate": 0.8571428571428571,
|
| 191 |
+
"policy_expert_regret": 1.128277287050918,
|
| 192 |
+
"policy_oracle_regret": 1.3004250411803906,
|
| 193 |
+
"policy_rollout_progress": 0.39018028597910326,
|
| 194 |
+
"policy_rollout_success_rate": 0.15384615384615385,
|
| 195 |
+
"restore_max_error": 3.948807716369629e-07
|
| 196 |
+
}
|
| 197 |
+
}
|
| 198 |
+
},
|
| 199 |
+
{
|
| 200 |
+
"seed": 2,
|
| 201 |
+
"path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/near_miss_policy_bc5/seed_2/policy_rollout_retrieval_residual_scale0p35_safe_types_margin0p60.json",
|
| 202 |
+
"num_groups": 575,
|
| 203 |
+
"selection_mode": "retrieval_residual",
|
| 204 |
+
"num_candidates": 16,
|
| 205 |
+
"candidate_sigma": 0.0,
|
| 206 |
+
"selection_margin": 0.6,
|
| 207 |
+
"field_optim_steps": 0,
|
| 208 |
+
"field_optim_step_size": 0.0,
|
| 209 |
+
"field_optim_trust_radius": 0.0,
|
| 210 |
+
"field_optim_l2_penalty": 0.0,
|
| 211 |
+
"retrieval_neighbors": 1,
|
| 212 |
+
"retrieval_metric": "raw",
|
| 213 |
+
"retrieval_type_min_success": 0.0,
|
| 214 |
+
"retrieval_residual_scale": 0.35,
|
| 215 |
+
"retrieval_residual_anchor": "expert",
|
| 216 |
+
"policy_rollout_success_rate": 0.3634782608695652,
|
| 217 |
+
"policy_rollout_progress": 0.5794369420074477,
|
| 218 |
+
"oracle_success_rate": 0.8765217391304347,
|
| 219 |
+
"action_mse_to_best": 0.41425420999284024,
|
| 220 |
+
"best_policy_val": {
|
| 221 |
+
"bc_loss": 0.11367896075050037,
|
| 222 |
+
"field_effect_loss": 0.009670218582161598,
|
| 223 |
+
"field_potential_loss": 0.2641640139950646,
|
| 224 |
+
"field_preference_loss": 0.5130490180518892,
|
| 225 |
+
"lattice_edges": 3833.3333333333335,
|
| 226 |
+
"progress_mae": 0.2021729110015763,
|
| 227 |
+
"rank_acc": 0.8333857821093665,
|
| 228 |
+
"rank_loss": 0.5130119257503085,
|
| 229 |
+
"regret_mae": 0.3958987047274907,
|
| 230 |
+
"success_accuracy": 0.8680730561415354,
|
| 231 |
+
"total_loss": 1.4394984311527677
|
| 232 |
+
},
|
| 233 |
+
"per_task": {
|
| 234 |
+
"LiftPegUpright-v1": {
|
| 235 |
+
"action_mse_to_best": 0.35361564260771655,
|
| 236 |
+
"expert_success_rate": 0.8229166666666666,
|
| 237 |
+
"num_groups": 96,
|
| 238 |
+
"oracle_success_rate": 0.9270833333333334,
|
| 239 |
+
"policy_expert_regret": 0.8470402403424183,
|
| 240 |
+
"policy_oracle_regret": 0.9533149556567272,
|
| 241 |
+
"policy_rollout_progress": 0.6469602342694998,
|
| 242 |
+
"policy_rollout_success_rate": 0.3333333333333333,
|
| 243 |
+
"restore_max_error": 3.5762786865234375e-07
|
| 244 |
+
},
|
| 245 |
+
"PickCube-v1": {
|
| 246 |
+
"action_mse_to_best": 0.33526025458493014,
|
| 247 |
+
"expert_success_rate": 0.9444444444444444,
|
| 248 |
+
"num_groups": 198,
|
| 249 |
+
"oracle_success_rate": 0.9595959595959596,
|
| 250 |
+
"policy_expert_regret": 0.995225173825718,
|
| 251 |
+
"policy_oracle_regret": 1.0023086243544264,
|
| 252 |
+
"policy_rollout_progress": 0.6313645500087678,
|
| 253 |
+
"policy_rollout_success_rate": 0.3282828282828283,
|
| 254 |
+
"restore_max_error": 4.76837158203125e-07
|
| 255 |
+
},
|
| 256 |
+
"PullCube-v1": {
|
| 257 |
+
"action_mse_to_best": 0.627604523466693,
|
| 258 |
+
"expert_success_rate": 0.24444444444444444,
|
| 259 |
+
"num_groups": 90,
|
| 260 |
+
"oracle_success_rate": 0.4666666666666667,
|
| 261 |
+
"policy_expert_regret": 0.2932497958204193,
|
| 262 |
+
"policy_oracle_regret": 0.5209218820005036,
|
| 263 |
+
"policy_rollout_progress": 0.32959150629877615,
|
| 264 |
+
"policy_rollout_success_rate": 0.2222222222222222,
|
| 265 |
+
"restore_max_error": 4.0978193283081055e-07
|
| 266 |
+
},
|
| 267 |
+
"PushCube-v1": {
|
| 268 |
+
"action_mse_to_best": 0.37755409141283225,
|
| 269 |
+
"expert_success_rate": 0.8514851485148515,
|
| 270 |
+
"num_groups": 101,
|
| 271 |
+
"oracle_success_rate": 1.0,
|
| 272 |
+
"policy_expert_regret": 0.380091660683698,
|
| 273 |
+
"policy_oracle_regret": 0.4527412828537497,
|
| 274 |
+
"policy_rollout_progress": 0.7848824795224879,
|
| 275 |
+
"policy_rollout_success_rate": 0.7623762376237624,
|
| 276 |
+
"restore_max_error": 4.76837158203125e-07
|
| 277 |
+
},
|
| 278 |
+
"StackCube-v1": {
|
| 279 |
+
"action_mse_to_best": 0.48055742558919723,
|
| 280 |
+
"expert_success_rate": 0.7666666666666667,
|
| 281 |
+
"num_groups": 90,
|
| 282 |
+
"oracle_success_rate": 0.9111111111111111,
|
| 283 |
+
"policy_expert_regret": 1.1538148525688383,
|
| 284 |
+
"policy_oracle_regret": 1.3028870176937846,
|
| 285 |
+
"policy_rollout_progress": 0.41246124737792544,
|
| 286 |
+
"policy_rollout_success_rate": 0.16666666666666666,
|
| 287 |
+
"restore_max_error": 4.76837158203125e-07
|
| 288 |
+
}
|
| 289 |
+
}
|
| 290 |
+
}
|
| 291 |
+
]
|
| 292 |
+
}
|
results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_scale0p35_safe_types_margin0p60_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_scale0p35_safe_types_margin0p60.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: 34.14% +/- 1.92%
|
| 11 |
+
Gain vs h=16 rank checkpoint: +4.41%
|
| 12 |
+
Mean progress: 56.08%
|
| 13 |
+
Mean action MSE to best: 0.394
|
| 14 |
+
|
| 15 |
+
| seed | mode | k | retrieval K | retrieval metric | residual anchor | min type success | residual scale | margin | sigma | opt steps | trust | success | progress | oracle | action MSE |
|
| 16 |
+
|---:|---|---:|---:|---|---|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|
|
| 17 |
+
| 0 | retrieval_residual | 16 | 1 | raw | expert | 0.00 | 0.35 | 0.600 | 0.00 | 0 | 0.00 | 33.22% | 54.67% | 85.74% | 0.381 |
|
| 18 |
+
| 1 | retrieval_residual | 16 | 1 | raw | expert | 0.00 | 0.35 | 0.600 | 0.00 | 0 | 0.00 | 32.87% | 55.61% | 86.96% | 0.388 |
|
| 19 |
+
| 2 | retrieval_residual | 16 | 1 | raw | expert | 0.00 | 0.35 | 0.600 | 0.00 | 0 | 0.00 | 36.35% | 57.94% | 87.65% | 0.414 |
|
results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_scale0p35_safe_types_margin1p00_summary.json
ADDED
|
@@ -0,0 +1,292 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 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_scale0p35_safe_types_margin1p00.json",
|
| 5 |
+
"num_completed": 3,
|
| 6 |
+
"baseline_h4_policy_success": 0.2967,
|
| 7 |
+
"baseline_h16_rank_checkpoint_success": 0.29739130434782607,
|
| 8 |
+
"mean_success": 0.3408695652173913,
|
| 9 |
+
"std_success": 0.018157054798105306,
|
| 10 |
+
"mean_progress": 0.5602468522778908,
|
| 11 |
+
"mean_action_mse_to_best": 0.3942014737751173,
|
| 12 |
+
"gain_vs_h4": 0.04416956521739129,
|
| 13 |
+
"gain_vs_h16_rank_checkpoint": 0.043478260869565244,
|
| 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_scale0p35_safe_types_margin1p00.json",
|
| 18 |
+
"num_groups": 575,
|
| 19 |
+
"selection_mode": "retrieval_residual",
|
| 20 |
+
"num_candidates": 16,
|
| 21 |
+
"candidate_sigma": 0.0,
|
| 22 |
+
"selection_margin": 1.0,
|
| 23 |
+
"field_optim_steps": 0,
|
| 24 |
+
"field_optim_step_size": 0.0,
|
| 25 |
+
"field_optim_trust_radius": 0.0,
|
| 26 |
+
"field_optim_l2_penalty": 0.0,
|
| 27 |
+
"retrieval_neighbors": 1,
|
| 28 |
+
"retrieval_metric": "raw",
|
| 29 |
+
"retrieval_type_min_success": 0.0,
|
| 30 |
+
"retrieval_residual_scale": 0.35,
|
| 31 |
+
"retrieval_residual_anchor": "expert",
|
| 32 |
+
"policy_rollout_success_rate": 0.33217391304347826,
|
| 33 |
+
"policy_rollout_progress": 0.5466740468091776,
|
| 34 |
+
"oracle_success_rate": 0.8573913043478261,
|
| 35 |
+
"action_mse_to_best": 0.38054896707039165,
|
| 36 |
+
"best_policy_val": {
|
| 37 |
+
"bc_loss": 0.13721593966086706,
|
| 38 |
+
"field_effect_loss": 0.009290305380192068,
|
| 39 |
+
"field_potential_loss": 0.2666468388504452,
|
| 40 |
+
"field_preference_loss": 0.5130573478009965,
|
| 41 |
+
"lattice_edges": 3833.3333333333335,
|
| 42 |
+
"progress_mae": 0.1933159919248687,
|
| 43 |
+
"rank_acc": 0.8265031774838766,
|
| 44 |
+
"rank_loss": 0.5130523675017886,
|
| 45 |
+
"regret_mae": 0.3756548762321472,
|
| 46 |
+
"success_accuracy": 0.8773836526605818,
|
| 47 |
+
"total_loss": 1.5581054819954767
|
| 48 |
+
},
|
| 49 |
+
"per_task": {
|
| 50 |
+
"LiftPegUpright-v1": {
|
| 51 |
+
"action_mse_to_best": 0.31969772514973566,
|
| 52 |
+
"expert_success_rate": 0.8865979381443299,
|
| 53 |
+
"num_groups": 97,
|
| 54 |
+
"oracle_success_rate": 0.9278350515463918,
|
| 55 |
+
"policy_expert_regret": 1.0792337135248578,
|
| 56 |
+
"policy_oracle_regret": 1.1056491912333006,
|
| 57 |
+
"policy_rollout_progress": 0.570563180084081,
|
| 58 |
+
"policy_rollout_success_rate": 0.2268041237113402,
|
| 59 |
+
"restore_max_error": 4.76837158203125e-07
|
| 60 |
+
},
|
| 61 |
+
"PickCube-v1": {
|
| 62 |
+
"action_mse_to_best": 0.30251795243328583,
|
| 63 |
+
"expert_success_rate": 0.9375,
|
| 64 |
+
"num_groups": 208,
|
| 65 |
+
"oracle_success_rate": 0.9471153846153846,
|
| 66 |
+
"policy_expert_regret": 0.9750573756531454,
|
| 67 |
+
"policy_oracle_regret": 0.9884802430796509,
|
| 68 |
+
"policy_rollout_progress": 0.6118246527890173,
|
| 69 |
+
"policy_rollout_success_rate": 0.3317307692307692,
|
| 70 |
+
"restore_max_error": 4.76837158203125e-07
|
| 71 |
+
},
|
| 72 |
+
"PullCube-v1": {
|
| 73 |
+
"action_mse_to_best": 0.607338386890176,
|
| 74 |
+
"expert_success_rate": 0.19480519480519481,
|
| 75 |
+
"num_groups": 77,
|
| 76 |
+
"oracle_success_rate": 0.36363636363636365,
|
| 77 |
+
"policy_expert_regret": 0.22117534107165773,
|
| 78 |
+
"policy_oracle_regret": 0.4382834872261125,
|
| 79 |
+
"policy_rollout_progress": 0.31418416151692247,
|
| 80 |
+
"policy_rollout_success_rate": 0.2077922077922078,
|
| 81 |
+
"restore_max_error": 4.76837158203125e-07
|
| 82 |
+
},
|
| 83 |
+
"PushCube-v1": {
|
| 84 |
+
"action_mse_to_best": 0.36476515277110516,
|
| 85 |
+
"expert_success_rate": 0.8617021276595744,
|
| 86 |
+
"num_groups": 94,
|
| 87 |
+
"oracle_success_rate": 0.9893617021276596,
|
| 88 |
+
"policy_expert_regret": 0.5719942746961371,
|
| 89 |
+
"policy_oracle_regret": 0.66845020953011,
|
| 90 |
+
"policy_rollout_progress": 0.682613620257124,
|
| 91 |
+
"policy_rollout_success_rate": 0.648936170212766,
|
| 92 |
+
"restore_max_error": 4.76837158203125e-07
|
| 93 |
+
},
|
| 94 |
+
"StackCube-v1": {
|
| 95 |
+
"action_mse_to_best": 0.4427097219070702,
|
| 96 |
+
"expert_success_rate": 0.6767676767676768,
|
| 97 |
+
"num_groups": 99,
|
| 98 |
+
"oracle_success_rate": 0.8585858585858586,
|
| 99 |
+
"policy_expert_regret": 0.9762899511390262,
|
| 100 |
+
"policy_oracle_regret": 1.1504495456965282,
|
| 101 |
+
"policy_rollout_progress": 0.4381369689498285,
|
| 102 |
+
"policy_rollout_success_rate": 0.23232323232323232,
|
| 103 |
+
"restore_max_error": 3.948807716369629e-07
|
| 104 |
+
}
|
| 105 |
+
}
|
| 106 |
+
},
|
| 107 |
+
{
|
| 108 |
+
"seed": 1,
|
| 109 |
+
"path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/near_miss_policy_bc5/seed_1/policy_rollout_retrieval_residual_scale0p35_safe_types_margin1p00.json",
|
| 110 |
+
"num_groups": 575,
|
| 111 |
+
"selection_mode": "retrieval_residual",
|
| 112 |
+
"num_candidates": 16,
|
| 113 |
+
"candidate_sigma": 0.0,
|
| 114 |
+
"selection_margin": 1.0,
|
| 115 |
+
"field_optim_steps": 0,
|
| 116 |
+
"field_optim_step_size": 0.0,
|
| 117 |
+
"field_optim_trust_radius": 0.0,
|
| 118 |
+
"field_optim_l2_penalty": 0.0,
|
| 119 |
+
"retrieval_neighbors": 1,
|
| 120 |
+
"retrieval_metric": "raw",
|
| 121 |
+
"retrieval_type_min_success": 0.0,
|
| 122 |
+
"retrieval_residual_scale": 0.35,
|
| 123 |
+
"retrieval_residual_anchor": "expert",
|
| 124 |
+
"policy_rollout_success_rate": 0.32869565217391306,
|
| 125 |
+
"policy_rollout_progress": 0.5560782212537267,
|
| 126 |
+
"oracle_success_rate": 0.8695652173913043,
|
| 127 |
+
"action_mse_to_best": 0.38778846145000145,
|
| 128 |
+
"best_policy_val": {
|
| 129 |
+
"bc_loss": 0.1315989200439718,
|
| 130 |
+
"field_effect_loss": 0.010110370814800262,
|
| 131 |
+
"field_potential_loss": 0.31546536915832096,
|
| 132 |
+
"field_preference_loss": 0.4960637440284093,
|
| 133 |
+
"lattice_edges": 3833.3333333333335,
|
| 134 |
+
"progress_mae": 0.19023226698239645,
|
| 135 |
+
"rank_acc": 0.8255469501018524,
|
| 136 |
+
"rank_loss": 0.4960531195004781,
|
| 137 |
+
"regret_mae": 0.42156532737943864,
|
| 138 |
+
"success_accuracy": 0.8807473679383596,
|
| 139 |
+
"total_loss": 1.561984618504842
|
| 140 |
+
},
|
| 141 |
+
"per_task": {
|
| 142 |
+
"LiftPegUpright-v1": {
|
| 143 |
+
"action_mse_to_best": 0.32557108186482064,
|
| 144 |
+
"expert_success_rate": 0.8584070796460177,
|
| 145 |
+
"num_groups": 113,
|
| 146 |
+
"oracle_success_rate": 0.9380530973451328,
|
| 147 |
+
"policy_expert_regret": 0.987694250535121,
|
| 148 |
+
"policy_oracle_regret": 1.066896405631462,
|
| 149 |
+
"policy_rollout_progress": 0.6352153201546289,
|
| 150 |
+
"policy_rollout_success_rate": 0.23008849557522124,
|
| 151 |
+
"restore_max_error": 4.76837158203125e-07
|
| 152 |
+
},
|
| 153 |
+
"PickCube-v1": {
|
| 154 |
+
"action_mse_to_best": 0.319557456076955,
|
| 155 |
+
"expert_success_rate": 0.9402173913043478,
|
| 156 |
+
"num_groups": 184,
|
| 157 |
+
"oracle_success_rate": 0.9456521739130435,
|
| 158 |
+
"policy_expert_regret": 1.0864933169467133,
|
| 159 |
+
"policy_oracle_regret": 1.0865173214216433,
|
| 160 |
+
"policy_rollout_progress": 0.5695780974954529,
|
| 161 |
+
"policy_rollout_success_rate": 0.2826086956521739,
|
| 162 |
+
"restore_max_error": 4.76837158203125e-07
|
| 163 |
+
},
|
| 164 |
+
"PullCube-v1": {
|
| 165 |
+
"action_mse_to_best": 0.5715608798751706,
|
| 166 |
+
"expert_success_rate": 0.25,
|
| 167 |
+
"num_groups": 76,
|
| 168 |
+
"oracle_success_rate": 0.40789473684210525,
|
| 169 |
+
"policy_expert_regret": 0.3506602184405844,
|
| 170 |
+
"policy_oracle_regret": 0.619953330042527,
|
| 171 |
+
"policy_rollout_progress": 0.24527493430497616,
|
| 172 |
+
"policy_rollout_success_rate": 0.13157894736842105,
|
| 173 |
+
"restore_max_error": 4.76837158203125e-07
|
| 174 |
+
},
|
| 175 |
+
"PushCube-v1": {
|
| 176 |
+
"action_mse_to_best": 0.3577534188222778,
|
| 177 |
+
"expert_success_rate": 0.8198198198198198,
|
| 178 |
+
"num_groups": 111,
|
| 179 |
+
"oracle_success_rate": 1.0,
|
| 180 |
+
"policy_expert_regret": 0.3483100165654947,
|
| 181 |
+
"policy_oracle_regret": 0.4142704948380187,
|
| 182 |
+
"policy_rollout_progress": 0.8019457213781975,
|
| 183 |
+
"policy_rollout_success_rate": 0.7837837837837838,
|
| 184 |
+
"restore_max_error": 4.76837158203125e-07
|
| 185 |
+
},
|
| 186 |
+
"StackCube-v1": {
|
| 187 |
+
"action_mse_to_best": 0.4861648879679186,
|
| 188 |
+
"expert_success_rate": 0.6923076923076923,
|
| 189 |
+
"num_groups": 91,
|
| 190 |
+
"oracle_success_rate": 0.8571428571428571,
|
| 191 |
+
"policy_expert_regret": 1.128277287050918,
|
| 192 |
+
"policy_oracle_regret": 1.3004250411803906,
|
| 193 |
+
"policy_rollout_progress": 0.39018028597910326,
|
| 194 |
+
"policy_rollout_success_rate": 0.15384615384615385,
|
| 195 |
+
"restore_max_error": 3.948807716369629e-07
|
| 196 |
+
}
|
| 197 |
+
}
|
| 198 |
+
},
|
| 199 |
+
{
|
| 200 |
+
"seed": 2,
|
| 201 |
+
"path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/near_miss_policy_bc5/seed_2/policy_rollout_retrieval_residual_scale0p35_safe_types_margin1p00.json",
|
| 202 |
+
"num_groups": 575,
|
| 203 |
+
"selection_mode": "retrieval_residual",
|
| 204 |
+
"num_candidates": 16,
|
| 205 |
+
"candidate_sigma": 0.0,
|
| 206 |
+
"selection_margin": 1.0,
|
| 207 |
+
"field_optim_steps": 0,
|
| 208 |
+
"field_optim_step_size": 0.0,
|
| 209 |
+
"field_optim_trust_radius": 0.0,
|
| 210 |
+
"field_optim_l2_penalty": 0.0,
|
| 211 |
+
"retrieval_neighbors": 1,
|
| 212 |
+
"retrieval_metric": "raw",
|
| 213 |
+
"retrieval_type_min_success": 0.0,
|
| 214 |
+
"retrieval_residual_scale": 0.35,
|
| 215 |
+
"retrieval_residual_anchor": "expert",
|
| 216 |
+
"policy_rollout_success_rate": 0.3617391304347826,
|
| 217 |
+
"policy_rollout_progress": 0.5779882887707682,
|
| 218 |
+
"oracle_success_rate": 0.8765217391304347,
|
| 219 |
+
"action_mse_to_best": 0.41426699280495877,
|
| 220 |
+
"best_policy_val": {
|
| 221 |
+
"bc_loss": 0.11367896075050037,
|
| 222 |
+
"field_effect_loss": 0.009670218582161598,
|
| 223 |
+
"field_potential_loss": 0.2641640139950646,
|
| 224 |
+
"field_preference_loss": 0.5130490180518892,
|
| 225 |
+
"lattice_edges": 3833.3333333333335,
|
| 226 |
+
"progress_mae": 0.2021729110015763,
|
| 227 |
+
"rank_acc": 0.8333857821093665,
|
| 228 |
+
"rank_loss": 0.5130119257503085,
|
| 229 |
+
"regret_mae": 0.3958987047274907,
|
| 230 |
+
"success_accuracy": 0.8680730561415354,
|
| 231 |
+
"total_loss": 1.4394984311527677
|
| 232 |
+
},
|
| 233 |
+
"per_task": {
|
| 234 |
+
"LiftPegUpright-v1": {
|
| 235 |
+
"action_mse_to_best": 0.35361564260771655,
|
| 236 |
+
"expert_success_rate": 0.8229166666666666,
|
| 237 |
+
"num_groups": 96,
|
| 238 |
+
"oracle_success_rate": 0.9270833333333334,
|
| 239 |
+
"policy_expert_regret": 0.8470402403424183,
|
| 240 |
+
"policy_oracle_regret": 0.9533149556567272,
|
| 241 |
+
"policy_rollout_progress": 0.6469602342694998,
|
| 242 |
+
"policy_rollout_success_rate": 0.3333333333333333,
|
| 243 |
+
"restore_max_error": 3.5762786865234375e-07
|
| 244 |
+
},
|
| 245 |
+
"PickCube-v1": {
|
| 246 |
+
"action_mse_to_best": 0.33526025458493014,
|
| 247 |
+
"expert_success_rate": 0.9444444444444444,
|
| 248 |
+
"num_groups": 198,
|
| 249 |
+
"oracle_success_rate": 0.9595959595959596,
|
| 250 |
+
"policy_expert_regret": 0.995225173825718,
|
| 251 |
+
"policy_oracle_regret": 1.0023086243544264,
|
| 252 |
+
"policy_rollout_progress": 0.6313645500087678,
|
| 253 |
+
"policy_rollout_success_rate": 0.3282828282828283,
|
| 254 |
+
"restore_max_error": 4.76837158203125e-07
|
| 255 |
+
},
|
| 256 |
+
"PullCube-v1": {
|
| 257 |
+
"action_mse_to_best": 0.6294899263315731,
|
| 258 |
+
"expert_success_rate": 0.24444444444444444,
|
| 259 |
+
"num_groups": 90,
|
| 260 |
+
"oracle_success_rate": 0.4666666666666667,
|
| 261 |
+
"policy_expert_regret": 0.29368426450699997,
|
| 262 |
+
"policy_oracle_regret": 0.5414197320808347,
|
| 263 |
+
"policy_rollout_progress": 0.3202047673295561,
|
| 264 |
+
"policy_rollout_success_rate": 0.2111111111111111,
|
| 265 |
+
"restore_max_error": 4.0978193283081055e-07
|
| 266 |
+
},
|
| 267 |
+
"PushCube-v1": {
|
| 268 |
+
"action_mse_to_best": 0.3759468028893565,
|
| 269 |
+
"expert_success_rate": 0.8514851485148515,
|
| 270 |
+
"num_groups": 101,
|
| 271 |
+
"oracle_success_rate": 1.0,
|
| 272 |
+
"policy_expert_regret": 0.37997452309816193,
|
| 273 |
+
"policy_oracle_regret": 0.45262414526821365,
|
| 274 |
+
"policy_rollout_progress": 0.784999617108024,
|
| 275 |
+
"policy_rollout_success_rate": 0.7623762376237624,
|
| 276 |
+
"restore_max_error": 4.76837158203125e-07
|
| 277 |
+
},
|
| 278 |
+
"StackCube-v1": {
|
| 279 |
+
"action_mse_to_best": 0.48055742558919723,
|
| 280 |
+
"expert_success_rate": 0.7666666666666667,
|
| 281 |
+
"num_groups": 90,
|
| 282 |
+
"oracle_success_rate": 0.9111111111111111,
|
| 283 |
+
"policy_expert_regret": 1.1538148525688383,
|
| 284 |
+
"policy_oracle_regret": 1.3028870176937846,
|
| 285 |
+
"policy_rollout_progress": 0.41246124737792544,
|
| 286 |
+
"policy_rollout_success_rate": 0.16666666666666666,
|
| 287 |
+
"restore_max_error": 4.76837158203125e-07
|
| 288 |
+
}
|
| 289 |
+
}
|
| 290 |
+
}
|
| 291 |
+
]
|
| 292 |
+
}
|
results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_scale0p35_safe_types_margin1p00_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_scale0p35_safe_types_margin1p00.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: 34.09% +/- 1.82%
|
| 11 |
+
Gain vs h=16 rank checkpoint: +4.35%
|
| 12 |
+
Mean progress: 56.02%
|
| 13 |
+
Mean action MSE to best: 0.394
|
| 14 |
+
|
| 15 |
+
| seed | mode | k | retrieval K | retrieval metric | residual anchor | min type success | residual scale | margin | sigma | opt steps | trust | success | progress | oracle | action MSE |
|
| 16 |
+
|---:|---|---:|---:|---|---|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|
|
| 17 |
+
| 0 | retrieval_residual | 16 | 1 | raw | expert | 0.00 | 0.35 | 1.000 | 0.00 | 0 | 0.00 | 33.22% | 54.67% | 85.74% | 0.381 |
|
| 18 |
+
| 1 | retrieval_residual | 16 | 1 | raw | expert | 0.00 | 0.35 | 1.000 | 0.00 | 0 | 0.00 | 32.87% | 55.61% | 86.96% | 0.388 |
|
| 19 |
+
| 2 | retrieval_residual | 16 | 1 | raw | expert | 0.00 | 0.35 | 1.000 | 0.00 | 0 | 0.00 | 36.17% | 57.80% | 87.65% | 0.414 |
|