anhtld commited on
Commit
771efec
·
verified ·
1 Parent(s): 12e1b40

Auto-sync: 2026-06-30 03:27:59 (part 2)

Browse files
results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_oraclek8trace_summary.json ADDED
@@ -0,0 +1,1653 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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_oraclek8trace.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_oraclek8trace.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_unique_tolerance": 1e-06,
59
+ "candidate_oracle_success_rate": 0.4226086956521739,
60
+ "candidate_oracle_progress": 0.6335964845250482,
61
+ "candidate_oracle_score": 1.056205180177222,
62
+ "candidate_oracle_regret": 0.7471866000800029,
63
+ "candidate_oracle_score_gain_over_selected": 0.1563077125147633,
64
+ "candidate_oracle_unique_count": 8.0,
65
+ "candidate_oracle_improvement_rate": 0.4643478260869565,
66
+ "candidate_oracle_selected_branch_success_rate": 0.3443478260869565,
67
+ "candidate_oracle_selected_branch_progress": 0.555549641262497,
68
+ "candidate_oracle_type_counts": {
69
+ "retrieval_residual_policy_residual": 280,
70
+ "retrieval_residual_residual_near_miss": 71,
71
+ "retrieval_residual_residual_near_miss+residual_wrong_gripper": 31,
72
+ "retrieval_residual_residual_no_op": 57,
73
+ "retrieval_residual_residual_no_op+residual_wrong_gripper": 31,
74
+ "retrieval_residual_residual_wrong_gripper": 105
75
+ },
76
+ "candidate_oracle_best_branch_rank": 2.897391304347826,
77
+ "candidate_oracle_best_branch_rank_counts": {
78
+ "1": 308,
79
+ "2": 55,
80
+ "3": 27,
81
+ "4": 25,
82
+ "5": 38,
83
+ "6": 33,
84
+ "7": 33,
85
+ "8": 56
86
+ },
87
+ "candidate_oracle_branch_success_rates": [
88
+ 0.3443478260869565,
89
+ 0.27652173913043476,
90
+ 0.2678260869565217,
91
+ 0.2608695652173913,
92
+ 0.24869565217391304,
93
+ 0.25043478260869567,
94
+ 0.2382608695652174,
95
+ 0.24347826086956523
96
+ ],
97
+ "candidate_oracle_branch_progress": [
98
+ 0.555549641262497,
99
+ 0.5051648186503546,
100
+ 0.4910595580579146,
101
+ 0.47653248145801785,
102
+ 0.47251200093199375,
103
+ 0.4663440991895597,
104
+ 0.45487754681674036,
105
+ 0.45191995076511216
106
+ ],
107
+ "candidate_oracle_branch_score_gains_over_selected": [
108
+ 0.0,
109
+ -0.1182109091669807,
110
+ -0.1410118218653066,
111
+ -0.16249542015614798,
112
+ -0.1786898131247448,
113
+ -0.18311858458140784,
114
+ -0.20675905063586392,
115
+ -0.20449925440282601
116
+ ],
117
+ "selected_residual_scale_counts": {
118
+ "0.35": 536,
119
+ "0.45": 35,
120
+ "0.4": 4
121
+ },
122
+ "policy_rollout_success_rate": 0.35130434782608694,
123
+ "policy_rollout_progress": 0.5579317773906919,
124
+ "oracle_success_rate": 0.8573913043478261,
125
+ "action_mse_to_best": 0.3930192190424904,
126
+ "best_policy_val": {
127
+ "bc_loss": 0.13721593966086706,
128
+ "field_effect_loss": 0.009290305380192068,
129
+ "field_potential_loss": 0.2666468388504452,
130
+ "field_preference_loss": 0.5130573478009965,
131
+ "lattice_edges": 3833.3333333333335,
132
+ "progress_mae": 0.1933159919248687,
133
+ "rank_acc": 0.8265031774838766,
134
+ "rank_loss": 0.5130523675017886,
135
+ "regret_mae": 0.3756548762321472,
136
+ "success_accuracy": 0.8773836526605818,
137
+ "total_loss": 1.5581054819954767
138
+ },
139
+ "per_task": {
140
+ "LiftPegUpright-v1": {
141
+ "action_mse_to_best": 0.3206862742465335,
142
+ "candidate_oracle_best_branch_rank": 2.3608247422680413,
143
+ "candidate_oracle_best_branch_rank_counts": {
144
+ "1": 66,
145
+ "2": 8,
146
+ "3": 1,
147
+ "5": 5,
148
+ "6": 6,
149
+ "7": 5,
150
+ "8": 6
151
+ },
152
+ "candidate_oracle_branch_progress": [
153
+ 0.5664646868238744,
154
+ 0.4579822413392903,
155
+ 0.4301517455233741,
156
+ 0.3987354222953934,
157
+ 0.42038861208969786,
158
+ 0.41464757089762344,
159
+ 0.39656632124763175,
160
+ 0.4040209738556872
161
+ ],
162
+ "candidate_oracle_branch_score_gains_over_selected": [
163
+ 0.0,
164
+ -0.19095667152060675,
165
+ -0.2084778892932479,
166
+ -0.2398942132893297,
167
+ -0.2491688565495088,
168
+ -0.23429133996521076,
169
+ -0.2523725886934811,
170
+ -0.23460866096093483
171
+ ],
172
+ "candidate_oracle_branch_scores": [
173
+ 0.7829595321846992,
174
+ 0.5920028598959913,
175
+ 0.5744816424305906,
176
+ 0.5430653192026099,
177
+ 0.5337906739453679,
178
+ 0.5486681894543245,
179
+ 0.5305869398043328,
180
+ 0.5483508707629037
181
+ ],
182
+ "candidate_oracle_branch_success_rates": [
183
+ 0.21649484536082475,
184
+ 0.13402061855670103,
185
+ 0.14432989690721648,
186
+ 0.14432989690721648,
187
+ 0.1134020618556701,
188
+ 0.13402061855670103,
189
+ 0.13402061855670103,
190
+ 0.14432989690721648
191
+ ],
192
+ "candidate_oracle_improvement_rate": 0.31958762886597936,
193
+ "candidate_oracle_progress": 0.5859934114611026,
194
+ "candidate_oracle_regret": 1.0799096815057636,
195
+ "candidate_oracle_restore_max_error": 4.76837158203125e-07,
196
+ "candidate_oracle_score": 0.8231068135229582,
197
+ "candidate_oracle_score_gain_over_selected": 0.040147280570158025,
198
+ "candidate_oracle_selected_branch_progress": 0.5664646868238744,
199
+ "candidate_oracle_selected_branch_success_rate": 0.21649484536082475,
200
+ "candidate_oracle_success_rate": 0.23711340206185566,
201
+ "candidate_oracle_type_counts": {
202
+ "retrieval_residual_policy_residual": 65,
203
+ "retrieval_residual_residual_near_miss": 14,
204
+ "retrieval_residual_residual_near_miss+residual_wrong_gripper": 6,
205
+ "retrieval_residual_residual_no_op": 1,
206
+ "retrieval_residual_residual_no_op+residual_wrong_gripper": 2,
207
+ "retrieval_residual_residual_wrong_gripper": 9
208
+ },
209
+ "candidate_oracle_unique_count": 8.0,
210
+ "expert_success_rate": 0.8865979381443299,
211
+ "num_groups": 97,
212
+ "oracle_success_rate": 0.9278350515463918,
213
+ "policy_expert_regret": 1.078290365866779,
214
+ "policy_oracle_regret": 1.1046864339678557,
215
+ "policy_rollout_progress": 0.571525937349526,
216
+ "policy_rollout_success_rate": 0.2268041237113402,
217
+ "restore_max_error": 2.384185791015625e-07
218
+ },
219
+ "PickCube-v1": {
220
+ "action_mse_to_best": 0.30273104519941485,
221
+ "candidate_oracle_best_branch_rank": 3.2163461538461537,
222
+ "candidate_oracle_best_branch_rank_counts": {
223
+ "1": 115,
224
+ "2": 13,
225
+ "4": 10,
226
+ "5": 14,
227
+ "6": 6,
228
+ "7": 18,
229
+ "8": 32
230
+ },
231
+ "candidate_oracle_branch_progress": [
232
+ 0.6204068260894229,
233
+ 0.5326739246374927,
234
+ 0.5306728121493227,
235
+ 0.5393201008510704,
236
+ 0.5654192451710025,
237
+ 0.5622782765735442,
238
+ 0.549470047896298,
239
+ 0.5632926820365426
240
+ ],
241
+ "candidate_oracle_branch_score_gains_over_selected": [
242
+ 0.0,
243
+ -0.20792520836515066,
244
+ -0.20031093787670565,
245
+ -0.18685595572997743,
246
+ -0.1126798873480696,
247
+ -0.11582085601716804,
248
+ -0.15747523865698335,
249
+ -0.10519106634176122
250
+ ],
251
+ "candidate_oracle_branch_scores": [
252
+ 0.9713683645509613,
253
+ 0.7634431554067235,
254
+ 0.771057427533938,
255
+ 0.7845124085433781,
256
+ 0.8586884759402332,
257
+ 0.855547507342775,
258
+ 0.813893124819375,
259
+ 0.8661772974211579
260
+ ],
261
+ "candidate_oracle_branch_success_rates": [
262
+ 0.35096153846153844,
263
+ 0.23076923076923078,
264
+ 0.2403846153846154,
265
+ 0.24519230769230768,
266
+ 0.2932692307692308,
267
+ 0.2932692307692308,
268
+ 0.2644230769230769,
269
+ 0.30288461538461536
270
+ ],
271
+ "candidate_oracle_improvement_rate": 0.44711538461538464,
272
+ "candidate_oracle_progress": 0.6512647587895537,
273
+ "candidate_oracle_regret": 0.8817324447714222,
274
+ "candidate_oracle_restore_max_error": 4.76837158203125e-07,
275
+ "candidate_oracle_score": 1.0503032203280152,
276
+ "candidate_oracle_score_gain_over_selected": 0.07893485469349588,
277
+ "candidate_oracle_selected_branch_progress": 0.6204068260894229,
278
+ "candidate_oracle_selected_branch_success_rate": 0.35096153846153844,
279
+ "candidate_oracle_success_rate": 0.39903846153846156,
280
+ "candidate_oracle_type_counts": {
281
+ "retrieval_residual_policy_residual": 101,
282
+ "retrieval_residual_residual_near_miss": 22,
283
+ "retrieval_residual_residual_near_miss+residual_wrong_gripper": 6,
284
+ "retrieval_residual_residual_no_op": 42,
285
+ "retrieval_residual_residual_wrong_gripper": 37
286
+ },
287
+ "candidate_oracle_unique_count": 8.0,
288
+ "expert_success_rate": 0.9375,
289
+ "num_groups": 208,
290
+ "oracle_success_rate": 0.9471153846153846,
291
+ "policy_expert_regret": 0.9543257619505032,
292
+ "policy_oracle_regret": 0.9677486293770087,
293
+ "policy_rollout_progress": 0.6181331895685827,
294
+ "policy_rollout_success_rate": 0.34615384615384615,
295
+ "restore_max_error": 2.2910535335540771e-07
296
+ },
297
+ "PullCube-v1": {
298
+ "action_mse_to_best": 0.6033564122853341,
299
+ "candidate_oracle_best_branch_rank": 3.4805194805194803,
300
+ "candidate_oracle_best_branch_rank_counts": {
301
+ "1": 21,
302
+ "2": 12,
303
+ "3": 10,
304
+ "4": 9,
305
+ "5": 6,
306
+ "6": 10,
307
+ "7": 5,
308
+ "8": 4
309
+ },
310
+ "candidate_oracle_branch_progress": [
311
+ 0.3112883431468062,
312
+ 0.31640808147608074,
313
+ 0.25566801384043,
314
+ 0.30286160906906806,
315
+ 0.293984739013113,
316
+ 0.3183850709684556,
317
+ 0.2759048884444229,
318
+ 0.2797975195450829
319
+ ],
320
+ "candidate_oracle_branch_score_gains_over_selected": [
321
+ 0.0,
322
+ 0.005119736515009752,
323
+ -0.1205553926690784,
324
+ -0.021413747043584702,
325
+ -0.04327763174358126,
326
+ 0.007096725653602995,
327
+ -0.07434449428329606,
328
+ -0.05746484751958255
329
+ ],
330
+ "candidate_oracle_branch_scores": [
331
+ 0.5190805509390141,
332
+ 0.5242002892682885,
333
+ 0.3985251566975728,
334
+ 0.4976668038742629,
335
+ 0.47580292083129483,
336
+ 0.5261772787606633,
337
+ 0.44473605727559173,
338
+ 0.46161570136326474
339
+ ],
340
+ "candidate_oracle_branch_success_rates": [
341
+ 0.2077922077922078,
342
+ 0.2077922077922078,
343
+ 0.14285714285714285,
344
+ 0.19480519480519481,
345
+ 0.18181818181818182,
346
+ 0.2077922077922078,
347
+ 0.16883116883116883,
348
+ 0.18181818181818182
349
+ ],
350
+ "candidate_oracle_improvement_rate": 0.7272727272727273,
351
+ "candidate_oracle_progress": 0.47178338712395784,
352
+ "candidate_oracle_regret": 0.1973970831166227,
353
+ "candidate_oracle_restore_max_error": 4.76837158203125e-07,
354
+ "candidate_oracle_score": 0.8354197507603215,
355
+ "candidate_oracle_score_gain_over_selected": 0.3163391990562925,
356
+ "candidate_oracle_selected_branch_progress": 0.3112883431468062,
357
+ "candidate_oracle_selected_branch_success_rate": 0.2077922077922078,
358
+ "candidate_oracle_success_rate": 0.36363636363636365,
359
+ "candidate_oracle_type_counts": {
360
+ "retrieval_residual_policy_residual": 17,
361
+ "retrieval_residual_residual_near_miss": 11,
362
+ "retrieval_residual_residual_near_miss+residual_wrong_gripper": 11,
363
+ "retrieval_residual_residual_no_op": 10,
364
+ "retrieval_residual_residual_no_op+residual_wrong_gripper": 5,
365
+ "retrieval_residual_residual_wrong_gripper": 23
366
+ },
367
+ "candidate_oracle_unique_count": 8.0,
368
+ "expert_success_rate": 0.19480519480519481,
369
+ "num_groups": 77,
370
+ "oracle_success_rate": 0.36363636363636365,
371
+ "policy_expert_regret": 0.2432419956728284,
372
+ "policy_oracle_regret": 0.41639802941459825,
373
+ "policy_rollout_progress": 0.3229174776722423,
374
+ "policy_rollout_success_rate": 0.22077922077922077,
375
+ "restore_max_error": 2.3096799850463867e-07
376
+ },
377
+ "PushCube-v1": {
378
+ "action_mse_to_best": 0.3675119048578942,
379
+ "candidate_oracle_best_branch_rank": 1.851063829787234,
380
+ "candidate_oracle_best_branch_rank_counts": {
381
+ "1": 66,
382
+ "2": 9,
383
+ "3": 7,
384
+ "4": 3,
385
+ "5": 3,
386
+ "6": 3,
387
+ "8": 3
388
+ },
389
+ "candidate_oracle_branch_progress": [
390
+ 0.6910807556928472,
391
+ 0.6980183738026213,
392
+ 0.6839556136029832,
393
+ 0.6286343743509435,
394
+ 0.5143204829160203,
395
+ 0.4695719106400267,
396
+ 0.5055010627875937,
397
+ 0.45669404369719485
398
+ ],
399
+ "candidate_oracle_branch_score_gains_over_selected": [
400
+ 0.0,
401
+ 0.017575917725867415,
402
+ -0.0071251405046341264,
403
+ -0.12627616397877958,
404
+ -0.3576113354969532,
405
+ -0.45555139618351104,
406
+ -0.3770690566681801,
407
+ -0.47906756052311433
408
+ ],
409
+ "candidate_oracle_branch_scores": [
410
+ 1.3506552237779537,
411
+ 1.3682311397600682,
412
+ 1.3435300816880895,
413
+ 1.2243790552020073,
414
+ 0.9930438871713395,
415
+ 0.8951038255336436,
416
+ 0.9735861691705724,
417
+ 0.8715876607184715
418
+ ],
419
+ "candidate_oracle_branch_success_rates": [
420
+ 0.6595744680851063,
421
+ 0.6702127659574468,
422
+ 0.6595744680851063,
423
+ 0.5957446808510638,
424
+ 0.4787234042553192,
425
+ 0.425531914893617,
426
+ 0.46808510638297873,
427
+ 0.4148936170212766
428
+ ],
429
+ "candidate_oracle_improvement_rate": 0.2978723404255319,
430
+ "candidate_oracle_progress": 0.8213912876362496,
431
+ "candidate_oracle_regret": 0.370098074065878,
432
+ "candidate_oracle_restore_max_error": 4.76837158203125e-07,
433
+ "candidate_oracle_score": 1.6299019259341219,
434
+ "candidate_oracle_score_gain_over_selected": 0.27924670405844426,
435
+ "candidate_oracle_selected_branch_progress": 0.6910807556928472,
436
+ "candidate_oracle_selected_branch_success_rate": 0.6595744680851063,
437
+ "candidate_oracle_success_rate": 0.8085106382978723,
438
+ "candidate_oracle_type_counts": {
439
+ "retrieval_residual_policy_residual": 63,
440
+ "retrieval_residual_residual_near_miss": 8,
441
+ "retrieval_residual_residual_near_miss+residual_wrong_gripper": 2,
442
+ "retrieval_residual_residual_no_op": 1,
443
+ "retrieval_residual_residual_no_op+residual_wrong_gripper": 2,
444
+ "retrieval_residual_residual_wrong_gripper": 18
445
+ },
446
+ "candidate_oracle_unique_count": 8.0,
447
+ "expert_success_rate": 0.8617021276595744,
448
+ "num_groups": 94,
449
+ "oracle_success_rate": 0.9893617021276596,
450
+ "policy_expert_regret": 0.5723979790159996,
451
+ "policy_oracle_regret": 0.6499486254884842,
452
+ "policy_rollout_progress": 0.6904769064264095,
453
+ "policy_rollout_success_rate": 0.6595744680851063,
454
+ "restore_max_error": 1.1920928955078125e-07
455
+ },
456
+ "StackCube-v1": {
457
+ "action_mse_to_best": 0.5142107287214862,
458
+ "candidate_oracle_best_branch_rank": 3.292929292929293,
459
+ "candidate_oracle_best_branch_rank_counts": {
460
+ "1": 40,
461
+ "2": 13,
462
+ "3": 9,
463
+ "4": 3,
464
+ "5": 10,
465
+ "6": 8,
466
+ "7": 5,
467
+ "8": 11
468
+ },
469
+ "candidate_oracle_branch_progress": [
470
+ 0.46988440222210354,
471
+ 0.4572946220034301,
472
+ 0.4674375444048583,
473
+ 0.4114978263775508,
474
+ 0.4275409279748647,
475
+ 0.42745203095855133,
476
+ 0.4044051513527379,
477
+ 0.3941960631295888
478
+ ],
479
+ "candidate_oracle_branch_score_gains_over_selected": [
480
+ 0.0,
481
+ -0.08329685122677774,
482
+ -0.09335594737168515,
483
+ -0.17959869871235856,
484
+ -0.1837576132531118,
485
+ -0.16364449232515663,
486
+ -0.20689339378867486,
487
+ -0.23730449920350855
488
+ ],
489
+ "candidate_oracle_branch_scores": [
490
+ 0.7325106648483661,
491
+ 0.649213813922622,
492
+ 0.63915471612203,
493
+ 0.5529119677916922,
494
+ 0.5487530491869859,
495
+ 0.5688661723726928,
496
+ 0.5256172725648591,
497
+ 0.4952061641396898
498
+ ],
499
+ "candidate_oracle_branch_success_rates": [
500
+ 0.26262626262626265,
501
+ 0.1919191919191919,
502
+ 0.1717171717171717,
503
+ 0.1414141414141414,
504
+ 0.12121212121212122,
505
+ 0.1414141414141414,
506
+ 0.12121212121212122,
507
+ 0.10101010101010101
508
+ ],
509
+ "candidate_oracle_improvement_rate": 0.5959595959595959,
510
+ "candidate_oracle_progress": 0.59066107086461,
511
+ "candidate_oracle_regret": 0.9241607380334778,
512
+ "candidate_oracle_restore_max_error": 4.76837158203125e-07,
513
+ "candidate_oracle_score": 0.9239944041979433,
514
+ "candidate_oracle_score_gain_over_selected": 0.19148373934957716,
515
+ "candidate_oracle_selected_branch_progress": 0.46988440222210354,
516
+ "candidate_oracle_selected_branch_success_rate": 0.26262626262626265,
517
+ "candidate_oracle_success_rate": 0.3333333333333333,
518
+ "candidate_oracle_type_counts": {
519
+ "retrieval_residual_policy_residual": 34,
520
+ "retrieval_residual_residual_near_miss": 16,
521
+ "retrieval_residual_residual_near_miss+residual_wrong_gripper": 6,
522
+ "retrieval_residual_residual_no_op": 3,
523
+ "retrieval_residual_residual_no_op+residual_wrong_gripper": 22,
524
+ "retrieval_residual_residual_wrong_gripper": 18
525
+ },
526
+ "candidate_oracle_unique_count": 8.0,
527
+ "expert_success_rate": 0.6767676767676768,
528
+ "num_groups": 99,
529
+ "oracle_success_rate": 0.8585858585858586,
530
+ "policy_expert_regret": 0.9052192963132955,
531
+ "policy_oracle_regret": 1.0644706299509665,
532
+ "policy_rollout_progress": 0.47506644102660095,
533
+ "policy_rollout_success_rate": 0.29292929292929293,
534
+ "restore_max_error": 2.1606683731079102e-07
535
+ }
536
+ }
537
+ },
538
+ {
539
+ "seed": 1,
540
+ "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_oraclek8trace.json",
541
+ "num_groups": 575,
542
+ "selection_mode": "retrieval_residual",
543
+ "num_candidates": 48,
544
+ "candidate_sigma": 0.0,
545
+ "selection_margin": 0.2,
546
+ "prepend_policy_candidate": false,
547
+ "field_optim_steps": 0,
548
+ "field_optim_step_size": 0.0,
549
+ "field_optim_trust_radius": 0.0,
550
+ "field_optim_l2_penalty": 0.0,
551
+ "retrieval_neighbors": 4,
552
+ "retrieval_metric": "raw",
553
+ "retrieval_type_min_success": 0.0,
554
+ "retrieval_type_success_bonus_scale": 0.0,
555
+ "retrieval_residual_consensus_penalty_scale": 0.0,
556
+ "retrieval_residual_min_source_progress": 0.0,
557
+ "retrieval_residual_min_source_advantage": -1000000000.0,
558
+ "retrieval_residual_source_progress_bonus_scale": 0.0,
559
+ "retrieval_residual_source_score_bonus_scale": 0.0,
560
+ "retrieval_residual_source_advantage_bonus_scale": 0.0,
561
+ "retrieval_residual_composite_l2_penalty_scale": 0.0,
562
+ "retrieval_residual_scale": 1.0,
563
+ "retrieval_residual_scales": [
564
+ 0.35,
565
+ 0.4,
566
+ 0.45
567
+ ],
568
+ "retrieval_residual_anchor": "expert",
569
+ "retrieval_residual_direction": "candidate_minus_anchor",
570
+ "retrieval_residual_reduce": "compose_mean_by_type",
571
+ "lattice_exclude_types": [
572
+ "residual_random_negative",
573
+ "residual_wrong_direction",
574
+ "residual_near_miss+residual_no_op"
575
+ ],
576
+ "candidate_type_bonuses": {
577
+ "residual_no_op": 0.03
578
+ },
579
+ "candidate_type_bonus_components": false,
580
+ "candidate_oracle_rollouts": 8,
581
+ "candidate_oracle_unique_tolerance": 1e-06,
582
+ "candidate_oracle_success_rate": 0.4156521739130435,
583
+ "candidate_oracle_progress": 0.6358434947061797,
584
+ "candidate_oracle_score": 1.0514956686192232,
585
+ "candidate_oracle_regret": 0.7618109145468992,
586
+ "candidate_oracle_score_gain_over_selected": 0.14767028470726118,
587
+ "candidate_oracle_unique_count": 8.0,
588
+ "candidate_oracle_improvement_rate": 0.46956521739130436,
589
+ "candidate_oracle_selected_branch_success_rate": 0.3391304347826087,
590
+ "candidate_oracle_selected_branch_progress": 0.5646949487568244,
591
+ "candidate_oracle_type_counts": {
592
+ "retrieval_residual_policy_residual": 285,
593
+ "retrieval_residual_residual_near_miss": 59,
594
+ "retrieval_residual_residual_near_miss+residual_wrong_gripper": 28,
595
+ "retrieval_residual_residual_no_op": 86,
596
+ "retrieval_residual_residual_no_op+residual_wrong_gripper": 27,
597
+ "retrieval_residual_residual_wrong_gripper": 90
598
+ },
599
+ "candidate_oracle_best_branch_rank": 2.7843478260869565,
600
+ "candidate_oracle_best_branch_rank_counts": {
601
+ "1": 305,
602
+ "2": 60,
603
+ "3": 29,
604
+ "4": 34,
605
+ "5": 39,
606
+ "6": 30,
607
+ "7": 46,
608
+ "8": 32
609
+ },
610
+ "candidate_oracle_branch_success_rates": [
611
+ 0.3391304347826087,
612
+ 0.2956521739130435,
613
+ 0.2782608695652174,
614
+ 0.271304347826087,
615
+ 0.24,
616
+ 0.23652173913043478,
617
+ 0.21913043478260869,
618
+ 0.24
619
+ ],
620
+ "candidate_oracle_branch_progress": [
621
+ 0.5646949487568244,
622
+ 0.5226921049554063,
623
+ 0.5042040664584214,
624
+ 0.4880521072949404,
625
+ 0.46781484017391567,
626
+ 0.4643760332176348,
627
+ 0.4498939143769119,
628
+ 0.45332161385862946
629
+ ],
630
+ "candidate_oracle_branch_score_gains_over_selected": [
631
+ 0.0,
632
+ -0.08548110417211834,
633
+ -0.12136044761945453,
634
+ -0.1444689283390408,
635
+ -0.19601054185434527,
636
+ -0.20292761081478522,
637
+ -0.234801034109424,
638
+ -0.2105037693258213
639
+ ],
640
+ "selected_residual_scale_counts": {
641
+ "0.35": 549,
642
+ "0.45": 23,
643
+ "0.4": 3
644
+ },
645
+ "policy_rollout_success_rate": 0.34782608695652173,
646
+ "policy_rollout_progress": 0.571580135051323,
647
+ "oracle_success_rate": 0.8695652173913043,
648
+ "action_mse_to_best": 0.39824521353549286,
649
+ "best_policy_val": {
650
+ "bc_loss": 0.1315989200439718,
651
+ "field_effect_loss": 0.010110370814800262,
652
+ "field_potential_loss": 0.31546536915832096,
653
+ "field_preference_loss": 0.4960637440284093,
654
+ "lattice_edges": 3833.3333333333335,
655
+ "progress_mae": 0.19023226698239645,
656
+ "rank_acc": 0.8255469501018524,
657
+ "rank_loss": 0.4960531195004781,
658
+ "regret_mae": 0.42156532737943864,
659
+ "success_accuracy": 0.8807473679383596,
660
+ "total_loss": 1.561984618504842
661
+ },
662
+ "per_task": {
663
+ "LiftPegUpright-v1": {
664
+ "action_mse_to_best": 0.3261080000104498,
665
+ "candidate_oracle_best_branch_rank": 2.433628318584071,
666
+ "candidate_oracle_best_branch_rank_counts": {
667
+ "1": 71,
668
+ "2": 7,
669
+ "3": 6,
670
+ "4": 2,
671
+ "5": 12,
672
+ "6": 6,
673
+ "7": 4,
674
+ "8": 5
675
+ },
676
+ "candidate_oracle_branch_progress": [
677
+ 0.6248437205774594,
678
+ 0.5124707165281329,
679
+ 0.5089766174818562,
680
+ 0.482778520446963,
681
+ 0.48673349236492564,
682
+ 0.4557350730210279,
683
+ 0.4575473020034554,
684
+ 0.4497275769183066
685
+ ],
686
+ "candidate_oracle_branch_score_gains_over_selected": [
687
+ 0.0,
688
+ -0.17431990604485031,
689
+ -0.1689644470415284,
690
+ -0.18631298734550983,
691
+ -0.19120757281780243,
692
+ -0.24875466525554657,
693
+ -0.23809287835538914,
694
+ -0.2282134909017951
695
+ ],
696
+ "candidate_oracle_branch_scores": [
697
+ 0.8460826586305568,
698
+ 0.671762751926363,
699
+ 0.6771182104022102,
700
+ 0.6597696708894409,
701
+ 0.6548750852852796,
702
+ 0.5973279933750102,
703
+ 0.6079897798795616,
704
+ 0.6178691698386606
705
+ ],
706
+ "candidate_oracle_branch_success_rates": [
707
+ 0.22123893805309736,
708
+ 0.1592920353982301,
709
+ 0.168141592920354,
710
+ 0.17699115044247787,
711
+ 0.168141592920354,
712
+ 0.1415929203539823,
713
+ 0.1504424778761062,
714
+ 0.168141592920354
715
+ ],
716
+ "candidate_oracle_improvement_rate": 0.37168141592920356,
717
+ "candidate_oracle_progress": 0.6674467935762575,
718
+ "candidate_oracle_regret": 0.9638451861333003,
719
+ "candidate_oracle_restore_max_error": 4.76837158203125e-07,
720
+ "candidate_oracle_score": 0.9683317493284698,
721
+ "candidate_oracle_score_gain_over_selected": 0.12224908885175148,
722
+ "candidate_oracle_selected_branch_progress": 0.6248437205774594,
723
+ "candidate_oracle_selected_branch_success_rate": 0.22123893805309736,
724
+ "candidate_oracle_success_rate": 0.3008849557522124,
725
+ "candidate_oracle_type_counts": {
726
+ "retrieval_residual_policy_residual": 70,
727
+ "retrieval_residual_residual_near_miss": 18,
728
+ "retrieval_residual_residual_near_miss+residual_wrong_gripper": 5,
729
+ "retrieval_residual_residual_no_op": 5,
730
+ "retrieval_residual_residual_no_op+residual_wrong_gripper": 1,
731
+ "retrieval_residual_residual_wrong_gripper": 14
732
+ },
733
+ "candidate_oracle_unique_count": 8.0,
734
+ "expert_success_rate": 0.8584070796460177,
735
+ "num_groups": 113,
736
+ "oracle_success_rate": 0.9380530973451328,
737
+ "policy_expert_regret": 0.9873958534635274,
738
+ "policy_oracle_regret": 1.0665980344061303,
739
+ "policy_rollout_progress": 0.6355136913799606,
740
+ "policy_rollout_success_rate": 0.23008849557522124,
741
+ "restore_max_error": 2.384185791015625e-07
742
+ },
743
+ "PickCube-v1": {
744
+ "action_mse_to_best": 0.32250052587250655,
745
+ "candidate_oracle_best_branch_rank": 2.9293478260869565,
746
+ "candidate_oracle_best_branch_rank_counts": {
747
+ "1": 89,
748
+ "2": 28,
749
+ "3": 5,
750
+ "4": 15,
751
+ "5": 9,
752
+ "6": 5,
753
+ "7": 20,
754
+ "8": 13
755
+ },
756
+ "candidate_oracle_branch_progress": [
757
+ 0.5768717735722337,
758
+ 0.5400854083102035,
759
+ 0.5428142085894132,
760
+ 0.5421088810898769,
761
+ 0.5325905407181658,
762
+ 0.5301836758710282,
763
+ 0.5217359400779494,
764
+ 0.5317791585365067
765
+ ],
766
+ "candidate_oracle_branch_score_gains_over_selected": [
767
+ 0.0,
768
+ -0.06396027815366245,
769
+ -0.05579669667286393,
770
+ -0.05106724109298185,
771
+ -0.0986290579236558,
772
+ -0.10647070521245831,
773
+ -0.13122278929728529,
774
+ -0.08313609365090403
775
+ ],
776
+ "candidate_oracle_branch_scores": [
777
+ 0.8757848170504946,
778
+ 0.8118245387449861,
779
+ 0.8199881216328915,
780
+ 0.8247175767420508,
781
+ 0.7771557581094701,
782
+ 0.7693141106536369,
783
+ 0.7445620270344712,
784
+ 0.792648723753898
785
+ ],
786
+ "candidate_oracle_branch_success_rates": [
787
+ 0.29891304347826086,
788
+ 0.2717391304347826,
789
+ 0.27717391304347827,
790
+ 0.2826086956521739,
791
+ 0.24456521739130435,
792
+ 0.2391304347826087,
793
+ 0.22282608695652173,
794
+ 0.2608695652173913
795
+ ],
796
+ "candidate_oracle_improvement_rate": 0.5163043478260869,
797
+ "candidate_oracle_progress": 0.6075394937728086,
798
+ "candidate_oracle_regret": 0.9779037512312441,
799
+ "candidate_oracle_restore_max_error": 4.76837158203125e-07,
800
+ "candidate_oracle_score": 0.960800363338026,
801
+ "candidate_oracle_score_gain_over_selected": 0.08501554508288593,
802
+ "candidate_oracle_selected_branch_progress": 0.5768717735722337,
803
+ "candidate_oracle_selected_branch_success_rate": 0.29891304347826086,
804
+ "candidate_oracle_success_rate": 0.3532608695652174,
805
+ "candidate_oracle_type_counts": {
806
+ "retrieval_residual_policy_residual": 81,
807
+ "retrieval_residual_residual_near_miss": 19,
808
+ "retrieval_residual_residual_near_miss+residual_wrong_gripper": 4,
809
+ "retrieval_residual_residual_no_op": 63,
810
+ "retrieval_residual_residual_no_op+residual_wrong_gripper": 2,
811
+ "retrieval_residual_residual_wrong_gripper": 15
812
+ },
813
+ "candidate_oracle_unique_count": 8.0,
814
+ "expert_success_rate": 0.9402173913043478,
815
+ "num_groups": 184,
816
+ "oracle_success_rate": 0.9456521739130435,
817
+ "policy_expert_regret": 1.0628609572152568,
818
+ "policy_oracle_regret": 1.0628849616901868,
819
+ "policy_rollout_progress": 0.5769061094008224,
820
+ "policy_rollout_success_rate": 0.29891304347826086,
821
+ "restore_max_error": 2.384185791015625e-07
822
+ },
823
+ "PullCube-v1": {
824
+ "action_mse_to_best": 0.5971616980080542,
825
+ "candidate_oracle_best_branch_rank": 3.460526315789474,
826
+ "candidate_oracle_best_branch_rank_counts": {
827
+ "1": 27,
828
+ "2": 7,
829
+ "3": 7,
830
+ "4": 8,
831
+ "5": 9,
832
+ "6": 7,
833
+ "7": 6,
834
+ "8": 5
835
+ },
836
+ "candidate_oracle_branch_progress": [
837
+ 0.277262910881913,
838
+ 0.19174682722721054,
839
+ 0.20855725684898635,
840
+ 0.19708002959132978,
841
+ 0.23502439875645856,
842
+ 0.19606232187475384,
843
+ 0.20893080834005223,
844
+ 0.23555370347713236
845
+ ],
846
+ "candidate_oracle_branch_score_gains_over_selected": [
847
+ 0.0,
848
+ -0.17762134320984937,
849
+ -0.14765302451269235,
850
+ -0.17228814488963076,
851
+ -0.09487009033756821,
852
+ -0.17330585139916366,
853
+ -0.1472794726172364,
854
+ -0.08118289120887455
855
+ ],
856
+ "candidate_oracle_branch_scores": [
857
+ 0.4483155424608604,
858
+ 0.27069419564826314,
859
+ 0.3006625200068811,
860
+ 0.2760273980123824,
861
+ 0.3534454513880375,
862
+ 0.27500969029580646,
863
+ 0.301036071497947,
864
+ 0.3671326508455534
865
+ ],
866
+ "candidate_oracle_branch_success_rates": [
867
+ 0.17105263157894737,
868
+ 0.07894736842105263,
869
+ 0.09210526315789473,
870
+ 0.07894736842105263,
871
+ 0.11842105263157894,
872
+ 0.07894736842105263,
873
+ 0.09210526315789473,
874
+ 0.13157894736842105
875
+ ],
876
+ "candidate_oracle_improvement_rate": 0.6447368421052632,
877
+ "candidate_oracle_progress": 0.41476133198624376,
878
+ "candidate_oracle_regret": 0.3187735332911344,
879
+ "candidate_oracle_restore_max_error": 4.76837158203125e-07,
880
+ "candidate_oracle_score": 0.7042350161967701,
881
+ "candidate_oracle_score_gain_over_selected": 0.25591947304967205,
882
+ "candidate_oracle_selected_branch_progress": 0.277262910881913,
883
+ "candidate_oracle_selected_branch_success_rate": 0.17105263157894737,
884
+ "candidate_oracle_success_rate": 0.2894736842105263,
885
+ "candidate_oracle_type_counts": {
886
+ "retrieval_residual_policy_residual": 23,
887
+ "retrieval_residual_residual_near_miss": 9,
888
+ "retrieval_residual_residual_near_miss+residual_wrong_gripper": 2,
889
+ "retrieval_residual_residual_no_op": 13,
890
+ "retrieval_residual_residual_no_op+residual_wrong_gripper": 8,
891
+ "retrieval_residual_residual_wrong_gripper": 21
892
+ },
893
+ "candidate_oracle_unique_count": 8.0,
894
+ "expert_success_rate": 0.25,
895
+ "num_groups": 76,
896
+ "oracle_success_rate": 0.40789473684210525,
897
+ "policy_expert_regret": 0.27669054590851855,
898
+ "policy_oracle_regret": 0.5001900389937586,
899
+ "policy_rollout_progress": 0.29926500726785316,
900
+ "policy_rollout_success_rate": 0.19736842105263158,
901
+ "restore_max_error": 2.3096799850463867e-07
902
+ },
903
+ "PushCube-v1": {
904
+ "action_mse_to_best": 0.3571035131677851,
905
+ "candidate_oracle_best_branch_rank": 1.6216216216216217,
906
+ "candidate_oracle_best_branch_rank_counts": {
907
+ "1": 89,
908
+ "2": 9,
909
+ "3": 3,
910
+ "4": 1,
911
+ "5": 1,
912
+ "6": 3,
913
+ "7": 3,
914
+ "8": 2
915
+ },
916
+ "candidate_oracle_branch_progress": [
917
+ 0.8178675184110263,
918
+ 0.8413939663955757,
919
+ 0.7509346078108022,
920
+ 0.7166112692237975,
921
+ 0.5917950617568987,
922
+ 0.6171278109153112,
923
+ 0.5479538067504093,
924
+ 0.5495762082639041
925
+ ],
926
+ "candidate_oracle_branch_score_gains_over_selected": [
927
+ 0.0,
928
+ 0.05055347568279988,
929
+ -0.1390049802558916,
930
+ -0.20936435447619842,
931
+ -0.46931569533305123,
932
+ -0.41695592465164427,
933
+ -0.5582019995461713,
934
+ -0.5565795972272083
935
+ ],
936
+ "candidate_oracle_branch_scores": [
937
+ 1.619669320212828,
938
+ 1.6702227952244046,
939
+ 1.480664337540532,
940
+ 1.410304962917491,
941
+ 1.1503536203154572,
942
+ 1.2027133965008967,
943
+ 1.0614673202639227,
944
+ 1.0630897217774176
945
+ ],
946
+ "candidate_oracle_branch_success_rates": [
947
+ 0.8018018018018018,
948
+ 0.8288288288288288,
949
+ 0.7297297297297297,
950
+ 0.6936936936936937,
951
+ 0.5585585585585585,
952
+ 0.5855855855855856,
953
+ 0.5135135135135135,
954
+ 0.5135135135135135
955
+ ],
956
+ "candidate_oracle_improvement_rate": 0.1981981981981982,
957
+ "candidate_oracle_progress": 0.9243861746949118,
958
+ "candidate_oracle_regret": 0.15669490638616923,
959
+ "candidate_oracle_restore_max_error": 4.76837158203125e-07,
960
+ "candidate_oracle_score": 1.8433050936138309,
961
+ "candidate_oracle_score_gain_over_selected": 0.2236357746092049,
962
+ "candidate_oracle_selected_branch_progress": 0.8178675184110263,
963
+ "candidate_oracle_selected_branch_success_rate": 0.8018018018018018,
964
+ "candidate_oracle_success_rate": 0.918918918918919,
965
+ "candidate_oracle_type_counts": {
966
+ "retrieval_residual_policy_residual": 83,
967
+ "retrieval_residual_residual_near_miss": 2,
968
+ "retrieval_residual_residual_near_miss+residual_wrong_gripper": 4,
969
+ "retrieval_residual_residual_no_op": 2,
970
+ "retrieval_residual_residual_no_op+residual_wrong_gripper": 2,
971
+ "retrieval_residual_residual_wrong_gripper": 18
972
+ },
973
+ "candidate_oracle_unique_count": 8.0,
974
+ "expert_success_rate": 0.8198198198198198,
975
+ "num_groups": 111,
976
+ "oracle_success_rate": 1.0,
977
+ "policy_expert_regret": 0.3456981908362191,
978
+ "policy_oracle_regret": 0.36286575796904863,
979
+ "policy_rollout_progress": 0.8263234312201405,
980
+ "policy_rollout_success_rate": 0.8108108108108109,
981
+ "restore_max_error": 1.7881393432617188e-07
982
+ },
983
+ "StackCube-v1": {
984
+ "action_mse_to_best": 0.5250320660543966,
985
+ "candidate_oracle_best_branch_rank": 3.78021978021978,
986
+ "candidate_oracle_best_branch_rank_counts": {
987
+ "1": 29,
988
+ "2": 9,
989
+ "3": 8,
990
+ "4": 8,
991
+ "5": 8,
992
+ "6": 9,
993
+ "7": 13,
994
+ "8": 7
995
+ },
996
+ "candidate_oracle_branch_progress": [
997
+ 0.3966216813404482,
998
+ 0.387863352894783,
999
+ 0.3661656381337197,
1000
+ 0.3495167212171869,
1001
+ 0.3565371737375364,
1002
+ 0.37980698916938277,
1003
+ 0.3767596558555142,
1004
+ 0.36360775851286375
1005
+ ],
1006
+ "candidate_oracle_branch_score_gains_over_selected": [
1007
+ 0.0,
1008
+ -0.10765942914800329,
1009
+ -0.15133516490459442,
1010
+ -0.17897309346513435,
1011
+ -0.14997461683802552,
1012
+ -0.1047267795919062,
1013
+ -0.11876312373103676,
1014
+ -0.13191501796245575
1015
+ ],
1016
+ "candidate_oracle_branch_scores": [
1017
+ 0.539478824197591,
1018
+ 0.431819396850827,
1019
+ 0.38814366011174173,
1020
+ 0.36050573220619786,
1021
+ 0.38950420670456937,
1022
+ 0.43475204411443774,
1023
+ 0.42071569981155815,
1024
+ 0.4075638024689077
1025
+ ],
1026
+ "candidate_oracle_branch_success_rates": [
1027
+ 0.14285714285714285,
1028
+ 0.04395604395604396,
1029
+ 0.02197802197802198,
1030
+ 0.01098901098901099,
1031
+ 0.03296703296703297,
1032
+ 0.054945054945054944,
1033
+ 0.04395604395604396,
1034
+ 0.04395604395604396
1035
+ ],
1036
+ "candidate_oracle_improvement_rate": 0.6813186813186813,
1037
+ "candidate_oracle_progress": 0.48651130006208526,
1038
+ "candidate_oracle_regret": 1.1821160051193866,
1039
+ "candidate_oracle_restore_max_error": 4.76837158203125e-07,
1040
+ "candidate_oracle_score": 0.6623354758862611,
1041
+ "candidate_oracle_score_gain_over_selected": 0.12285665316241128,
1042
+ "candidate_oracle_selected_branch_progress": 0.3966216813404482,
1043
+ "candidate_oracle_selected_branch_success_rate": 0.14285714285714285,
1044
+ "candidate_oracle_success_rate": 0.17582417582417584,
1045
+ "candidate_oracle_type_counts": {
1046
+ "retrieval_residual_policy_residual": 28,
1047
+ "retrieval_residual_residual_near_miss": 11,
1048
+ "retrieval_residual_residual_near_miss+residual_wrong_gripper": 13,
1049
+ "retrieval_residual_residual_no_op": 3,
1050
+ "retrieval_residual_residual_no_op+residual_wrong_gripper": 14,
1051
+ "retrieval_residual_residual_wrong_gripper": 22
1052
+ },
1053
+ "candidate_oracle_unique_count": 8.0,
1054
+ "expert_success_rate": 0.6923076923076923,
1055
+ "num_groups": 91,
1056
+ "oracle_success_rate": 0.8571428571428571,
1057
+ "policy_expert_regret": 1.1328262393618678,
1058
+ "policy_oracle_regret": 1.292487030664643,
1059
+ "policy_rollout_progress": 0.39811829649485075,
1060
+ "policy_rollout_success_rate": 0.15384615384615385,
1061
+ "restore_max_error": 1.9371509552001953e-07
1062
+ }
1063
+ }
1064
+ },
1065
+ {
1066
+ "seed": 2,
1067
+ "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_oraclek8trace.json",
1068
+ "num_groups": 575,
1069
+ "selection_mode": "retrieval_residual",
1070
+ "num_candidates": 48,
1071
+ "candidate_sigma": 0.0,
1072
+ "selection_margin": 0.2,
1073
+ "prepend_policy_candidate": false,
1074
+ "field_optim_steps": 0,
1075
+ "field_optim_step_size": 0.0,
1076
+ "field_optim_trust_radius": 0.0,
1077
+ "field_optim_l2_penalty": 0.0,
1078
+ "retrieval_neighbors": 4,
1079
+ "retrieval_metric": "raw",
1080
+ "retrieval_type_min_success": 0.0,
1081
+ "retrieval_type_success_bonus_scale": 0.0,
1082
+ "retrieval_residual_consensus_penalty_scale": 0.0,
1083
+ "retrieval_residual_min_source_progress": 0.0,
1084
+ "retrieval_residual_min_source_advantage": -1000000000.0,
1085
+ "retrieval_residual_source_progress_bonus_scale": 0.0,
1086
+ "retrieval_residual_source_score_bonus_scale": 0.0,
1087
+ "retrieval_residual_source_advantage_bonus_scale": 0.0,
1088
+ "retrieval_residual_composite_l2_penalty_scale": 0.0,
1089
+ "retrieval_residual_scale": 1.0,
1090
+ "retrieval_residual_scales": [
1091
+ 0.35,
1092
+ 0.4,
1093
+ 0.45
1094
+ ],
1095
+ "retrieval_residual_anchor": "expert",
1096
+ "retrieval_residual_direction": "candidate_minus_anchor",
1097
+ "retrieval_residual_reduce": "compose_mean_by_type",
1098
+ "lattice_exclude_types": [
1099
+ "residual_random_negative",
1100
+ "residual_wrong_direction",
1101
+ "residual_near_miss+residual_no_op"
1102
+ ],
1103
+ "candidate_type_bonuses": {
1104
+ "residual_no_op": 0.03
1105
+ },
1106
+ "candidate_type_bonus_components": false,
1107
+ "candidate_oracle_rollouts": 8,
1108
+ "candidate_oracle_unique_tolerance": 1e-06,
1109
+ "candidate_oracle_success_rate": 0.4539130434782609,
1110
+ "candidate_oracle_progress": 0.6635855871827706,
1111
+ "candidate_oracle_score": 1.1174986306610315,
1112
+ "candidate_oracle_regret": 0.7082039576120999,
1113
+ "candidate_oracle_score_gain_over_selected": 0.1770689690598975,
1114
+ "candidate_oracle_unique_count": 8.0,
1115
+ "candidate_oracle_improvement_rate": 0.4469565217391304,
1116
+ "candidate_oracle_selected_branch_success_rate": 0.3617391304347826,
1117
+ "candidate_oracle_selected_branch_progress": 0.57869053049364,
1118
+ "candidate_oracle_type_counts": {
1119
+ "retrieval_residual_policy_residual": 302,
1120
+ "retrieval_residual_residual_near_miss": 71,
1121
+ "retrieval_residual_residual_near_miss+residual_wrong_gripper": 40,
1122
+ "retrieval_residual_residual_no_op": 39,
1123
+ "retrieval_residual_residual_no_op+residual_wrong_gripper": 28,
1124
+ "retrieval_residual_residual_wrong_gripper": 95
1125
+ },
1126
+ "candidate_oracle_best_branch_rank": 2.8817391304347826,
1127
+ "candidate_oracle_best_branch_rank_counts": {
1128
+ "1": 318,
1129
+ "2": 47,
1130
+ "3": 32,
1131
+ "4": 28,
1132
+ "5": 25,
1133
+ "6": 24,
1134
+ "7": 40,
1135
+ "8": 61
1136
+ },
1137
+ "candidate_oracle_branch_success_rates": [
1138
+ 0.3617391304347826,
1139
+ 0.30782608695652175,
1140
+ 0.28347826086956524,
1141
+ 0.2678260869565217,
1142
+ 0.28347826086956524,
1143
+ 0.26608695652173914,
1144
+ 0.24347826086956523,
1145
+ 0.2678260869565217
1146
+ ],
1147
+ "candidate_oracle_branch_progress": [
1148
+ 0.57869053049364,
1149
+ 0.5326541016923021,
1150
+ 0.4994642052961432,
1151
+ 0.4885194452115051,
1152
+ 0.5028948070594798,
1153
+ 0.4870382411737481,
1154
+ 0.4685426553889461,
1155
+ 0.4900894078061632
1156
+ ],
1157
+ "candidate_oracle_branch_score_gains_over_selected": [
1158
+ 0.0,
1159
+ -0.0999494722428853,
1160
+ -0.15748719506613587,
1161
+ -0.18408412898931167,
1162
+ -0.1540565932477298,
1163
+ -0.18730446287876237,
1164
+ -0.22840874377800072,
1165
+ -0.18251416606909554
1166
+ ],
1167
+ "selected_residual_scale_counts": {
1168
+ "0.35": 546,
1169
+ "0.45": 25,
1170
+ "0.4": 4
1171
+ },
1172
+ "policy_rollout_success_rate": 0.37043478260869567,
1173
+ "policy_rollout_progress": 0.5852133044125383,
1174
+ "oracle_success_rate": 0.8765217391304347,
1175
+ "action_mse_to_best": 0.42779136531621864,
1176
+ "best_policy_val": {
1177
+ "bc_loss": 0.11367896075050037,
1178
+ "field_effect_loss": 0.009670218582161598,
1179
+ "field_potential_loss": 0.2641640139950646,
1180
+ "field_preference_loss": 0.5130490180518892,
1181
+ "lattice_edges": 3833.3333333333335,
1182
+ "progress_mae": 0.2021729110015763,
1183
+ "rank_acc": 0.8333857821093665,
1184
+ "rank_loss": 0.5130119257503085,
1185
+ "regret_mae": 0.3958987047274907,
1186
+ "success_accuracy": 0.8680730561415354,
1187
+ "total_loss": 1.4394984311527677
1188
+ },
1189
+ "per_task": {
1190
+ "LiftPegUpright-v1": {
1191
+ "action_mse_to_best": 0.35442226944239036,
1192
+ "candidate_oracle_best_branch_rank": 2.6979166666666665,
1193
+ "candidate_oracle_best_branch_rank_counts": {
1194
+ "1": 58,
1195
+ "2": 5,
1196
+ "3": 6,
1197
+ "4": 2,
1198
+ "5": 6,
1199
+ "6": 4,
1200
+ "7": 9,
1201
+ "8": 6
1202
+ },
1203
+ "candidate_oracle_branch_progress": [
1204
+ 0.6328173445848128,
1205
+ 0.5623508949453632,
1206
+ 0.5506054647266865,
1207
+ 0.540301852238675,
1208
+ 0.51264580215017,
1209
+ 0.49654898621762794,
1210
+ 0.4735308294184506,
1211
+ 0.47048968200882274
1212
+ ],
1213
+ "candidate_oracle_branch_score_gains_over_selected": [
1214
+ 0.0,
1215
+ -0.14338311475391188,
1216
+ -0.14471187985812625,
1217
+ -0.14459882273028293,
1218
+ -0.22433820956697068,
1219
+ -0.24043502348164716,
1220
+ -0.29470317841817933,
1221
+ -0.3081609949780007
1222
+ ],
1223
+ "candidate_oracle_branch_scores": [
1224
+ 0.934900677918146,
1225
+ 0.7915175616120299,
1226
+ 0.7901887980600198,
1227
+ 0.790301852238675,
1228
+ 0.7105624688168367,
1229
+ 0.6944656528842946,
1230
+ 0.6401974960851172,
1231
+ 0.6267396820088228
1232
+ ],
1233
+ "candidate_oracle_branch_success_rates": [
1234
+ 0.3020833333333333,
1235
+ 0.22916666666666666,
1236
+ 0.23958333333333334,
1237
+ 0.25,
1238
+ 0.19791666666666666,
1239
+ 0.19791666666666666,
1240
+ 0.16666666666666666,
1241
+ 0.15625
1242
+ ],
1243
+ "candidate_oracle_improvement_rate": 0.3958333333333333,
1244
+ "candidate_oracle_progress": 0.6717962672313055,
1245
+ "candidate_oracle_regret": 0.8763955893615881,
1246
+ "candidate_oracle_restore_max_error": 4.76837158203125e-07,
1247
+ "candidate_oracle_score": 1.057212933897972,
1248
+ "candidate_oracle_score_gain_over_selected": 0.12231225349629919,
1249
+ "candidate_oracle_selected_branch_progress": 0.6328173445848128,
1250
+ "candidate_oracle_selected_branch_success_rate": 0.3020833333333333,
1251
+ "candidate_oracle_success_rate": 0.3854166666666667,
1252
+ "candidate_oracle_type_counts": {
1253
+ "retrieval_residual_policy_residual": 57,
1254
+ "retrieval_residual_residual_near_miss": 14,
1255
+ "retrieval_residual_residual_near_miss+residual_wrong_gripper": 6,
1256
+ "retrieval_residual_residual_no_op": 1,
1257
+ "retrieval_residual_residual_no_op+residual_wrong_gripper": 2,
1258
+ "retrieval_residual_residual_wrong_gripper": 16
1259
+ },
1260
+ "candidate_oracle_unique_count": 8.0,
1261
+ "expert_success_rate": 0.8229166666666666,
1262
+ "num_groups": 96,
1263
+ "oracle_success_rate": 0.9270833333333334,
1264
+ "policy_expert_regret": 0.8475987246880928,
1265
+ "policy_oracle_regret": 0.9538775502393643,
1266
+ "policy_rollout_progress": 0.6463976396868626,
1267
+ "policy_rollout_success_rate": 0.3333333333333333,
1268
+ "restore_max_error": 2.384185791015625e-07
1269
+ },
1270
+ "PickCube-v1": {
1271
+ "action_mse_to_best": 0.3390917487705898,
1272
+ "candidate_oracle_best_branch_rank": 2.797979797979798,
1273
+ "candidate_oracle_best_branch_rank_counts": {
1274
+ "1": 118,
1275
+ "2": 11,
1276
+ "3": 9,
1277
+ "4": 12,
1278
+ "5": 6,
1279
+ "6": 7,
1280
+ "7": 13,
1281
+ "8": 22
1282
+ },
1283
+ "candidate_oracle_branch_progress": [
1284
+ 0.6357145610387729,
1285
+ 0.5654954555152794,
1286
+ 0.5676934397262003,
1287
+ 0.5512516928949591,
1288
+ 0.5556206395049965,
1289
+ 0.5504601229866496,
1290
+ 0.5560675962555288,
1291
+ 0.5865509132525385
1292
+ ],
1293
+ "candidate_oracle_branch_score_gains_over_selected": [
1294
+ 0.0,
1295
+ -0.12577466203859358,
1296
+ -0.1185261725796142,
1297
+ -0.1652709481899034,
1298
+ -0.13059897068417584,
1299
+ -0.15091100278677363,
1300
+ -0.13015201408416033,
1301
+ -0.08956768937089075
1302
+ ],
1303
+ "candidate_oracle_branch_scores": [
1304
+ 0.9740983994226112,
1305
+ 0.8483237383435621,
1306
+ 0.8555722276049883,
1307
+ 0.8088274504707167,
1308
+ 0.8434994273837844,
1309
+ 0.8231873957139223,
1310
+ 0.8439463841343167,
1311
+ 0.8845307112323365
1312
+ ],
1313
+ "candidate_oracle_branch_success_rates": [
1314
+ 0.3383838383838384,
1315
+ 0.2828282828282828,
1316
+ 0.2878787878787879,
1317
+ 0.25757575757575757,
1318
+ 0.2878787878787879,
1319
+ 0.2727272727272727,
1320
+ 0.2878787878787879,
1321
+ 0.29797979797979796
1322
+ ],
1323
+ "candidate_oracle_improvement_rate": 0.40404040404040403,
1324
+ "candidate_oracle_progress": 0.6581640573067948,
1325
+ "candidate_oracle_regret": 0.9300545716018538,
1326
+ "candidate_oracle_restore_max_error": 4.76837158203125e-07,
1327
+ "candidate_oracle_score": 1.0319014310441685,
1328
+ "candidate_oracle_score_gain_over_selected": 0.05780303117941425,
1329
+ "candidate_oracle_selected_branch_progress": 0.6357145610387729,
1330
+ "candidate_oracle_selected_branch_success_rate": 0.3383838383838384,
1331
+ "candidate_oracle_success_rate": 0.37373737373737376,
1332
+ "candidate_oracle_type_counts": {
1333
+ "retrieval_residual_policy_residual": 114,
1334
+ "retrieval_residual_residual_near_miss": 21,
1335
+ "retrieval_residual_residual_near_miss+residual_wrong_gripper": 6,
1336
+ "retrieval_residual_residual_no_op": 23,
1337
+ "retrieval_residual_residual_no_op+residual_wrong_gripper": 4,
1338
+ "retrieval_residual_residual_wrong_gripper": 30
1339
+ },
1340
+ "candidate_oracle_unique_count": 8.0,
1341
+ "expert_success_rate": 0.9444444444444444,
1342
+ "num_groups": 198,
1343
+ "oracle_success_rate": 0.9595959595959596,
1344
+ "policy_expert_regret": 0.9807323978442435,
1345
+ "policy_oracle_regret": 0.9878158483729519,
1346
+ "policy_rollout_progress": 0.6357563158892321,
1347
+ "policy_rollout_success_rate": 0.3383838383838384,
1348
+ "restore_max_error": 2.384185791015625e-07
1349
+ },
1350
+ "PullCube-v1": {
1351
+ "action_mse_to_best": 0.6408373928732343,
1352
+ "candidate_oracle_best_branch_rank": 3.6777777777777776,
1353
+ "candidate_oracle_best_branch_rank_counts": {
1354
+ "1": 34,
1355
+ "2": 12,
1356
+ "3": 7,
1357
+ "4": 5,
1358
+ "5": 3,
1359
+ "6": 3,
1360
+ "7": 9,
1361
+ "8": 17
1362
+ },
1363
+ "candidate_oracle_branch_progress": [
1364
+ 0.3270835054900393,
1365
+ 0.22817256613634526,
1366
+ 0.26075001319663393,
1367
+ 0.23498956476752128,
1368
+ 0.25593739666251675,
1369
+ 0.28468709347459176,
1370
+ 0.22975221652951505,
1371
+ 0.32866308391094207
1372
+ ],
1373
+ "candidate_oracle_branch_score_gains_over_selected": [
1374
+ 0.0,
1375
+ -0.21002205076834393,
1376
+ -0.13300016071233484,
1377
+ -0.1809828286110941,
1378
+ -0.1378127765841782,
1379
+ -0.07572974594206446,
1380
+ -0.19733128673914407,
1381
+ 0.012690688773161835
1382
+ ],
1383
+ "candidate_oracle_branch_scores": [
1384
+ 0.5493057277122615,
1385
+ 0.3392836772474564,
1386
+ 0.4163055687521895,
1387
+ 0.3683228981008546,
1388
+ 0.4114929522180723,
1389
+ 0.47357598236348064,
1390
+ 0.3519744387517373,
1391
+ 0.5619964172442754
1392
+ ],
1393
+ "candidate_oracle_branch_success_rates": [
1394
+ 0.2222222222222222,
1395
+ 0.1111111111111111,
1396
+ 0.15555555555555556,
1397
+ 0.13333333333333333,
1398
+ 0.15555555555555556,
1399
+ 0.18888888888888888,
1400
+ 0.12222222222222222,
1401
+ 0.23333333333333334
1402
+ ],
1403
+ "candidate_oracle_improvement_rate": 0.6222222222222222,
1404
+ "candidate_oracle_progress": 0.4730292216978139,
1405
+ "candidate_oracle_regret": 0.30388461639069847,
1406
+ "candidate_oracle_restore_max_error": 4.76837158203125e-07,
1407
+ "candidate_oracle_score": 0.8508069994755917,
1408
+ "candidate_oracle_score_gain_over_selected": 0.3015012719151047,
1409
+ "candidate_oracle_selected_branch_progress": 0.3270835054900393,
1410
+ "candidate_oracle_selected_branch_success_rate": 0.2222222222222222,
1411
+ "candidate_oracle_success_rate": 0.37777777777777777,
1412
+ "candidate_oracle_type_counts": {
1413
+ "retrieval_residual_policy_residual": 31,
1414
+ "retrieval_residual_residual_near_miss": 21,
1415
+ "retrieval_residual_residual_near_miss+residual_wrong_gripper": 18,
1416
+ "retrieval_residual_residual_no_op": 6,
1417
+ "retrieval_residual_residual_no_op+residual_wrong_gripper": 8,
1418
+ "retrieval_residual_residual_wrong_gripper": 6
1419
+ },
1420
+ "candidate_oracle_unique_count": 8.0,
1421
+ "expert_success_rate": 0.24444444444444444,
1422
+ "num_groups": 90,
1423
+ "oracle_success_rate": 0.4666666666666667,
1424
+ "policy_expert_regret": 0.27485709413195863,
1425
+ "policy_oracle_regret": 0.5026470268076183,
1426
+ "policy_rollout_progress": 0.3366318846604372,
1427
+ "policy_rollout_success_rate": 0.23333333333333334,
1428
+ "restore_max_error": 2.384185791015625e-07
1429
+ },
1430
+ "PushCube-v1": {
1431
+ "action_mse_to_best": 0.3970492488824495,
1432
+ "candidate_oracle_best_branch_rank": 1.7326732673267327,
1433
+ "candidate_oracle_best_branch_rank_counts": {
1434
+ "1": 80,
1435
+ "2": 8,
1436
+ "3": 2,
1437
+ "4": 1,
1438
+ "5": 2,
1439
+ "6": 1,
1440
+ "7": 3,
1441
+ "8": 4
1442
+ },
1443
+ "candidate_oracle_branch_progress": [
1444
+ 0.8090742473850155,
1445
+ 0.8171506218685962,
1446
+ 0.6438372831533451,
1447
+ 0.6425737000928067,
1448
+ 0.6700455909318263,
1449
+ 0.5998270139835848,
1450
+ 0.5534984702521032,
1451
+ 0.5664018404365766
1452
+ ],
1453
+ "candidate_oracle_branch_score_gains_over_selected": [
1454
+ 0.0,
1455
+ 0.01797736399244554,
1456
+ -0.3434547836532687,
1457
+ -0.354619365074847,
1458
+ -0.28754350985630905,
1459
+ -0.4369700070064847,
1460
+ -0.542704491479562,
1461
+ -0.5099991384414163
1462
+ ],
1463
+ "candidate_oracle_branch_scores": [
1464
+ 1.6011534553058078,
1465
+ 1.6191308198883982,
1466
+ 1.257698669291959,
1467
+ 1.2465340961324107,
1468
+ 1.3136099473674698,
1469
+ 1.1641834496271493,
1470
+ 1.058448965301608,
1471
+ 1.0911543156841013
1472
+ ],
1473
+ "candidate_oracle_branch_success_rates": [
1474
+ 0.7920792079207921,
1475
+ 0.801980198019802,
1476
+ 0.6138613861386139,
1477
+ 0.6039603960396039,
1478
+ 0.6435643564356436,
1479
+ 0.5643564356435643,
1480
+ 0.504950495049505,
1481
+ 0.5247524752475248
1482
+ ],
1483
+ "candidate_oracle_improvement_rate": 0.2079207920792079,
1484
+ "candidate_oracle_progress": 0.9358177499310805,
1485
+ "candidate_oracle_regret": 0.1334891807619888,
1486
+ "candidate_oracle_restore_max_error": 4.76837158203125e-07,
1487
+ "candidate_oracle_score": 1.866510819238011,
1488
+ "candidate_oracle_score_gain_over_selected": 0.26535736245684105,
1489
+ "candidate_oracle_selected_branch_progress": 0.8090742473850155,
1490
+ "candidate_oracle_selected_branch_success_rate": 0.7920792079207921,
1491
+ "candidate_oracle_success_rate": 0.9306930693069307,
1492
+ "candidate_oracle_type_counts": {
1493
+ "retrieval_residual_policy_residual": 73,
1494
+ "retrieval_residual_residual_near_miss": 2,
1495
+ "retrieval_residual_residual_near_miss+residual_wrong_gripper": 2,
1496
+ "retrieval_residual_residual_no_op": 4,
1497
+ "retrieval_residual_residual_no_op+residual_wrong_gripper": 2,
1498
+ "retrieval_residual_residual_wrong_gripper": 18
1499
+ },
1500
+ "candidate_oracle_unique_count": 8.0,
1501
+ "expert_success_rate": 0.8514851485148515,
1502
+ "num_groups": 101,
1503
+ "oracle_success_rate": 1.0,
1504
+ "policy_expert_regret": 0.3425388766987489,
1505
+ "policy_oracle_regret": 0.39744112810285964,
1506
+ "policy_rollout_progress": 0.8104796639763483,
1507
+ "policy_rollout_success_rate": 0.7920792079207921,
1508
+ "restore_max_error": 2.384185791015625e-07
1509
+ },
1510
+ "StackCube-v1": {
1511
+ "action_mse_to_best": 0.5226443493117888,
1512
+ "candidate_oracle_best_branch_rank": 3.7555555555555555,
1513
+ "candidate_oracle_best_branch_rank_counts": {
1514
+ "1": 28,
1515
+ "2": 11,
1516
+ "3": 8,
1517
+ "4": 8,
1518
+ "5": 8,
1519
+ "6": 9,
1520
+ "7": 6,
1521
+ "8": 12
1522
+ },
1523
+ "candidate_oracle_branch_progress": [
1524
+ 0.3885676932003763,
1525
+ 0.41393976228104695,
1526
+ 0.37150472866164314,
1527
+ 0.37592070500055946,
1528
+ 0.43587400052282543,
1529
+ 0.4131423869066768,
1530
+ 0.41411820203065874,
1531
+ 0.3745672861735026
1532
+ ],
1533
+ "candidate_oracle_branch_score_gains_over_selected": [
1534
+ 0.0,
1535
+ -0.01907237370808919,
1536
+ -0.07261852125326793,
1537
+ -0.07931365552875731,
1538
+ 0.0028618620501624213,
1539
+ -0.04209197229809231,
1540
+ -0.05222727010647456,
1541
+ -0.08066707203785578
1542
+ ],
1543
+ "candidate_oracle_branch_scores": [
1544
+ 0.5219010265337096,
1545
+ 0.5028286511699359,
1546
+ 0.4492825064394209,
1547
+ 0.44258737166722617,
1548
+ 0.5247628894117143,
1549
+ 0.47980905357334347,
1550
+ 0.46967375758621427,
1551
+ 0.44123395284016925
1552
+ ],
1553
+ "candidate_oracle_branch_success_rates": [
1554
+ 0.13333333333333333,
1555
+ 0.08888888888888889,
1556
+ 0.07777777777777778,
1557
+ 0.06666666666666667,
1558
+ 0.08888888888888889,
1559
+ 0.06666666666666667,
1560
+ 0.05555555555555555,
1561
+ 0.06666666666666667
1562
+ ],
1563
+ "candidate_oracle_improvement_rate": 0.6888888888888889,
1564
+ "candidate_oracle_progress": 0.5518062770366668,
1565
+ "candidate_oracle_regret": 1.0900052348772684,
1566
+ "candidate_oracle_restore_max_error": 4.76837158203125e-07,
1567
+ "candidate_oracle_score": 0.7962507214811113,
1568
+ "candidate_oracle_score_gain_over_selected": 0.274349695775244,
1569
+ "candidate_oracle_selected_branch_progress": 0.3885676932003763,
1570
+ "candidate_oracle_selected_branch_success_rate": 0.13333333333333333,
1571
+ "candidate_oracle_success_rate": 0.24444444444444444,
1572
+ "candidate_oracle_type_counts": {
1573
+ "retrieval_residual_policy_residual": 27,
1574
+ "retrieval_residual_residual_near_miss": 13,
1575
+ "retrieval_residual_residual_near_miss+residual_wrong_gripper": 8,
1576
+ "retrieval_residual_residual_no_op": 5,
1577
+ "retrieval_residual_residual_no_op+residual_wrong_gripper": 12,
1578
+ "retrieval_residual_residual_wrong_gripper": 25
1579
+ },
1580
+ "candidate_oracle_unique_count": 8.0,
1581
+ "expert_success_rate": 0.7666666666666667,
1582
+ "num_groups": 90,
1583
+ "oracle_success_rate": 0.9111111111111111,
1584
+ "policy_expert_regret": 1.1750739713509877,
1585
+ "policy_oracle_regret": 1.3371944083107843,
1586
+ "policy_rollout_progress": 0.4045378933350245,
1587
+ "policy_rollout_success_rate": 0.14444444444444443,
1588
+ "restore_max_error": 2.086162567138672e-07
1589
+ }
1590
+ }
1591
+ }
1592
+ ],
1593
+ "candidate_oracle_rollouts": 8,
1594
+ "candidate_oracle_unique_tolerance": 1e-06,
1595
+ "mean_candidate_oracle_success_rate": 0.43072463768115943,
1596
+ "mean_candidate_oracle_progress": 0.6443418554713328,
1597
+ "mean_candidate_oracle_score": 1.0750664931524923,
1598
+ "mean_candidate_oracle_regret": 0.7390671574130007,
1599
+ "mean_candidate_oracle_score_gain_over_selected": 0.16034898876064066,
1600
+ "mean_candidate_oracle_unique_count": 8.0,
1601
+ "mean_candidate_oracle_improvement_rate": 0.4602898550724638,
1602
+ "mean_candidate_oracle_selected_branch_success_rate": 0.34840579710144925,
1603
+ "mean_candidate_oracle_selected_branch_progress": 0.5663117068376537,
1604
+ "candidate_oracle_type_counts": {
1605
+ "retrieval_residual_policy_residual": 867,
1606
+ "retrieval_residual_residual_near_miss": 201,
1607
+ "retrieval_residual_residual_near_miss+residual_wrong_gripper": 99,
1608
+ "retrieval_residual_residual_no_op": 182,
1609
+ "retrieval_residual_residual_no_op+residual_wrong_gripper": 86,
1610
+ "retrieval_residual_residual_wrong_gripper": 290
1611
+ },
1612
+ "mean_candidate_oracle_best_branch_rank": 2.8544927536231883,
1613
+ "candidate_oracle_best_branch_rank_counts": {
1614
+ "1": 931,
1615
+ "2": 162,
1616
+ "3": 88,
1617
+ "4": 87,
1618
+ "5": 102,
1619
+ "6": 87,
1620
+ "7": 119,
1621
+ "8": 149
1622
+ },
1623
+ "mean_candidate_oracle_branch_success_rates": [
1624
+ 0.34840579710144925,
1625
+ 0.29333333333333333,
1626
+ 0.27652173913043476,
1627
+ 0.26666666666666666,
1628
+ 0.2573913043478261,
1629
+ 0.2510144927536232,
1630
+ 0.2336231884057971,
1631
+ 0.25043478260869567
1632
+ ],
1633
+ "mean_candidate_oracle_branch_progress": [
1634
+ 0.5663117068376537,
1635
+ 0.520170341766021,
1636
+ 0.4982426099374931,
1637
+ 0.48436801132148777,
1638
+ 0.4810738827217964,
1639
+ 0.4725861245269809,
1640
+ 0.45777137219419944,
1641
+ 0.4651103241433016
1642
+ ],
1643
+ "mean_candidate_oracle_branch_score_gains_over_selected": [
1644
+ 0.0,
1645
+ -0.10121382852732812,
1646
+ -0.139953154850299,
1647
+ -0.16368282582816682,
1648
+ -0.17625231607560662,
1649
+ -0.1911168860916518,
1650
+ -0.22332294284109622,
1651
+ -0.19917239659924763
1652
+ ]
1653
+ }
results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_oraclek8trace_summary.md ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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_oraclek8trace.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 unique tolerance: 1e-06
16
+ Candidate-oracle success: 43.07%
17
+ Candidate-oracle progress: 64.43%
18
+ Candidate-oracle gain over selected branch: +0.160
19
+ Candidate-oracle unique count: 8.00
20
+ Candidate-oracle improvement rate: 46.03%
21
+ Candidate-oracle best branch rank: 2.85
22
+ Candidate-oracle best branch rank counts: {'1': 931, '2': 162, '3': 88, '4': 87, '5': 102, '6': 87, '7': 119, '8': 149}
23
+ Candidate-oracle branch success by rank: 34.84%, 29.33%, 27.65%, 26.67%, 25.74%, 25.10%, 23.36%, 25.04%
24
+ Candidate-oracle branch gain by rank: +0.000, -0.101, -0.140, -0.164, -0.176, -0.191, -0.223, -0.199
25
+
26
+ | 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 |
27
+ |---:|---|---:|---|---:|---|---|---|---|---:|---:|---:|---:|---:|---:|---:|---|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|
28
+ | 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% | 42.26% | +0.156 | 0.393 |
29
+ | 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% | 41.57% | +0.148 | 0.398 |
30
+ | 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% | 45.39% | +0.177 | 0.428 |