anhtld commited on
Commit
27451f8
·
verified ·
1 Parent(s): bee5d92

auto-sync 2026-07-02T14:58:13Z workspace (part 7)

Browse files
workspace/results/h16_transport_field_reground_fieldonly_k6clean_dropnoopwg_b12_v1_besttransport_margin0p00_k6_srcscore001_metric_taskrel_summary.json ADDED
@@ -0,0 +1,481 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "run_root": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs",
3
+ "objective": "transport_field_reground_fieldonly_k6clean_dropnoopwg_b12_v1",
4
+ "out_name": "policy_rollout_fieldonly_k6clean_dropnoopwg_b12_v1_besttransport_margin0p00_k6_srcscore001_metric_taskrel.json",
5
+ "num_completed": 3,
6
+ "baseline_h4_policy_success": 0.2967,
7
+ "baseline_h16_rank_checkpoint_success": 0.29739130434782607,
8
+ "mean_success": 0.37217391304347824,
9
+ "std_success": 0.0165902469811643,
10
+ "mean_progress": 0.5848761159114108,
11
+ "mean_action_mse_to_best": 0.5116538966142073,
12
+ "gain_vs_h4": 0.07547391304347822,
13
+ "gain_vs_h16_rank_checkpoint": 0.07478260869565218,
14
+ "rows": [
15
+ {
16
+ "seed": 0,
17
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/transport_field_reground_fieldonly_k6clean_dropnoopwg_b12_v1/seed_0/policy_rollout_fieldonly_k6clean_dropnoopwg_b12_v1_besttransport_margin0p00_k6_srcscore001_metric_taskrel.json",
18
+ "num_groups": 575,
19
+ "selection_mode": "retrieval_residual",
20
+ "num_candidates": 48,
21
+ "candidate_sigma": 0.0,
22
+ "proposal_types": [],
23
+ "proposal_lattice_types": [],
24
+ "selection_margin": 0.0,
25
+ "prepend_policy_candidate": false,
26
+ "field_optim_steps": 0,
27
+ "field_optim_step_size": 0.0,
28
+ "field_optim_trust_radius": 0.0,
29
+ "field_optim_l2_penalty": 0.0,
30
+ "retrieval_neighbors": 6,
31
+ "retrieval_metric": "task_relative",
32
+ "retrieval_type_min_success": 0.0,
33
+ "retrieval_type_success_bonus_scale": 0.0,
34
+ "retrieval_residual_consensus_penalty_scale": 0.0,
35
+ "retrieval_residual_min_source_progress": 0.0,
36
+ "retrieval_residual_min_source_advantage": -1000000000.0,
37
+ "retrieval_residual_source_progress_bonus_scale": 0.0,
38
+ "retrieval_residual_source_score_bonus_scale": 0.01,
39
+ "retrieval_residual_source_advantage_bonus_scale": 0.0,
40
+ "retrieval_residual_composite_l2_penalty_scale": 0.0,
41
+ "retrieval_residual_scale": 1.0,
42
+ "retrieval_residual_scales": [
43
+ 0.35,
44
+ 0.4,
45
+ 0.45
46
+ ],
47
+ "retrieval_residual_anchor": "expert",
48
+ "retrieval_residual_direction": "candidate_minus_anchor",
49
+ "retrieval_residual_reduce": "compose_mean_by_type",
50
+ "retrieval_residual_challenger_types": [],
51
+ "retrieval_residual_challenger_scales": [],
52
+ "retrieval_residual_challenger_margin": 0.0,
53
+ "retrieval_residual_challenger_type_margins": {},
54
+ "retrieval_residual_challenger_tasks": [],
55
+ "retrieval_residual_challenger_type_tasks": {},
56
+ "lattice_exclude_types": [
57
+ "residual_random_negative",
58
+ "residual_wrong_direction",
59
+ "residual_near_miss+residual_no_op",
60
+ "residual_no_op+residual_wrong_gripper"
61
+ ],
62
+ "lattice_exclude_type_tasks": {},
63
+ "candidate_type_bonuses": {},
64
+ "candidate_type_bonuses_by_task": {},
65
+ "candidate_type_bonus_components": false,
66
+ "candidate_oracle_rollouts": 0,
67
+ "candidate_oracle_unique_tolerance": 1e-06,
68
+ "candidate_oracle_success_rate": null,
69
+ "candidate_oracle_progress": null,
70
+ "candidate_oracle_score": null,
71
+ "candidate_oracle_regret": null,
72
+ "candidate_oracle_score_gain_over_selected": null,
73
+ "candidate_oracle_unique_count": null,
74
+ "candidate_oracle_improvement_rate": null,
75
+ "candidate_oracle_selected_branch_success_rate": null,
76
+ "candidate_oracle_selected_branch_progress": null,
77
+ "candidate_oracle_type_counts": {},
78
+ "candidate_oracle_best_branch_rank": null,
79
+ "candidate_oracle_best_branch_rank_counts": {},
80
+ "candidate_oracle_branch_success_rates": [],
81
+ "candidate_oracle_branch_progress": [],
82
+ "candidate_oracle_branch_score_gains_over_selected": [],
83
+ "selected_residual_scale_counts": {
84
+ "0.35": 319,
85
+ "0.45": 186,
86
+ "0.4": 70
87
+ },
88
+ "policy_rollout_success_rate": 0.3739130434782609,
89
+ "policy_rollout_progress": 0.5859811081583409,
90
+ "oracle_success_rate": 0.8573913043478261,
91
+ "action_mse_to_best": 0.4960412207715537,
92
+ "best_policy_val": {
93
+ "bc_loss": 0.3445456624031067,
94
+ "field_effect_loss": 0.009637858977334367,
95
+ "field_potential_loss": 0.28480341533819836,
96
+ "field_preference_loss": 0.5244891477955712,
97
+ "lattice_edges": 3833.3333333333335,
98
+ "progress_mae": 0.19576937291357252,
99
+ "proposal_loss": 0.0,
100
+ "rank_acc": 0.8192069795396593,
101
+ "rank_loss": 0.524484740363227,
102
+ "regret_mae": 0.42191632754272884,
103
+ "success_accuracy": 0.8693191342883639,
104
+ "total_loss": 0.261704473859734,
105
+ "transport_field_candidates": 0.0,
106
+ "transport_field_edges": 0.0,
107
+ "transport_field_loss": 0.0,
108
+ "transport_field_potential_loss": 0.0,
109
+ "transport_field_preference_loss": 0.0,
110
+ "transport_field_rank_acc": 0.0
111
+ },
112
+ "per_task": {
113
+ "LiftPegUpright-v1": {
114
+ "action_mse_to_best": 0.3438506922636604,
115
+ "expert_success_rate": 0.8865979381443299,
116
+ "num_groups": 97,
117
+ "oracle_success_rate": 0.9278350515463918,
118
+ "policy_expert_regret": 1.078284000766646,
119
+ "policy_oracle_regret": 1.104699478475089,
120
+ "policy_rollout_progress": 0.5715128928422928,
121
+ "policy_rollout_success_rate": 0.2268041237113402,
122
+ "restore_max_error": 2.384185791015625e-07
123
+ },
124
+ "PickCube-v1": {
125
+ "action_mse_to_best": 0.46656056665457213,
126
+ "expert_success_rate": 0.9375,
127
+ "num_groups": 208,
128
+ "oracle_success_rate": 0.9471153846153846,
129
+ "policy_expert_regret": 0.908604594744527,
130
+ "policy_oracle_regret": 0.9152711687985664,
131
+ "policy_rollout_progress": 0.641764496300871,
132
+ "policy_rollout_success_rate": 0.375,
133
+ "restore_max_error": 2.384185791015625e-07
134
+ },
135
+ "PullCube-v1": {
136
+ "action_mse_to_best": 0.632325976119413,
137
+ "expert_success_rate": 0.19480519480519481,
138
+ "num_groups": 77,
139
+ "oracle_success_rate": 0.36363636363636365,
140
+ "policy_expert_regret": 0.2681577504588315,
141
+ "policy_oracle_regret": 0.4569121295066139,
142
+ "policy_rollout_progress": 0.3286020585771215,
143
+ "policy_rollout_success_rate": 0.22077922077922077,
144
+ "restore_max_error": 2.3096799850463867e-07
145
+ },
146
+ "PushCube-v1": {
147
+ "action_mse_to_best": 0.46895278633591975,
148
+ "expert_success_rate": 0.8617021276595744,
149
+ "num_groups": 94,
150
+ "oracle_success_rate": 0.9893617021276596,
151
+ "policy_expert_regret": 0.4561503028615992,
152
+ "policy_oracle_regret": 0.5146354934636582,
153
+ "policy_rollout_progress": 0.7513219533448524,
154
+ "policy_rollout_success_rate": 0.7340425531914894,
155
+ "restore_max_error": 2.384185791015625e-07
156
+ },
157
+ "StackCube-v1": {
158
+ "action_mse_to_best": 0.6268174227590513,
159
+ "expert_success_rate": 0.6767676767676768,
160
+ "num_groups": 99,
161
+ "oracle_success_rate": 0.8585858585858586,
162
+ "policy_expert_regret": 0.8581753356589211,
163
+ "policy_oracle_regret": 1.0285984300302737,
164
+ "policy_rollout_progress": 0.5261489821202827,
165
+ "policy_rollout_success_rate": 0.29292929292929293,
166
+ "restore_max_error": 2.384185791015625e-07
167
+ }
168
+ }
169
+ },
170
+ {
171
+ "seed": 1,
172
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/transport_field_reground_fieldonly_k6clean_dropnoopwg_b12_v1/seed_1/policy_rollout_fieldonly_k6clean_dropnoopwg_b12_v1_besttransport_margin0p00_k6_srcscore001_metric_taskrel.json",
173
+ "num_groups": 575,
174
+ "selection_mode": "retrieval_residual",
175
+ "num_candidates": 48,
176
+ "candidate_sigma": 0.0,
177
+ "proposal_types": [],
178
+ "proposal_lattice_types": [],
179
+ "selection_margin": 0.0,
180
+ "prepend_policy_candidate": false,
181
+ "field_optim_steps": 0,
182
+ "field_optim_step_size": 0.0,
183
+ "field_optim_trust_radius": 0.0,
184
+ "field_optim_l2_penalty": 0.0,
185
+ "retrieval_neighbors": 6,
186
+ "retrieval_metric": "task_relative",
187
+ "retrieval_type_min_success": 0.0,
188
+ "retrieval_type_success_bonus_scale": 0.0,
189
+ "retrieval_residual_consensus_penalty_scale": 0.0,
190
+ "retrieval_residual_min_source_progress": 0.0,
191
+ "retrieval_residual_min_source_advantage": -1000000000.0,
192
+ "retrieval_residual_source_progress_bonus_scale": 0.0,
193
+ "retrieval_residual_source_score_bonus_scale": 0.01,
194
+ "retrieval_residual_source_advantage_bonus_scale": 0.0,
195
+ "retrieval_residual_composite_l2_penalty_scale": 0.0,
196
+ "retrieval_residual_scale": 1.0,
197
+ "retrieval_residual_scales": [
198
+ 0.35,
199
+ 0.4,
200
+ 0.45
201
+ ],
202
+ "retrieval_residual_anchor": "expert",
203
+ "retrieval_residual_direction": "candidate_minus_anchor",
204
+ "retrieval_residual_reduce": "compose_mean_by_type",
205
+ "retrieval_residual_challenger_types": [],
206
+ "retrieval_residual_challenger_scales": [],
207
+ "retrieval_residual_challenger_margin": 0.0,
208
+ "retrieval_residual_challenger_type_margins": {},
209
+ "retrieval_residual_challenger_tasks": [],
210
+ "retrieval_residual_challenger_type_tasks": {},
211
+ "lattice_exclude_types": [
212
+ "residual_random_negative",
213
+ "residual_wrong_direction",
214
+ "residual_near_miss+residual_no_op",
215
+ "residual_no_op+residual_wrong_gripper"
216
+ ],
217
+ "lattice_exclude_type_tasks": {},
218
+ "candidate_type_bonuses": {},
219
+ "candidate_type_bonuses_by_task": {},
220
+ "candidate_type_bonus_components": false,
221
+ "candidate_oracle_rollouts": 0,
222
+ "candidate_oracle_unique_tolerance": 1e-06,
223
+ "candidate_oracle_success_rate": null,
224
+ "candidate_oracle_progress": null,
225
+ "candidate_oracle_score": null,
226
+ "candidate_oracle_regret": null,
227
+ "candidate_oracle_score_gain_over_selected": null,
228
+ "candidate_oracle_unique_count": null,
229
+ "candidate_oracle_improvement_rate": null,
230
+ "candidate_oracle_selected_branch_success_rate": null,
231
+ "candidate_oracle_selected_branch_progress": null,
232
+ "candidate_oracle_type_counts": {},
233
+ "candidate_oracle_best_branch_rank": null,
234
+ "candidate_oracle_best_branch_rank_counts": {},
235
+ "candidate_oracle_branch_success_rates": [],
236
+ "candidate_oracle_branch_progress": [],
237
+ "candidate_oracle_branch_score_gains_over_selected": [],
238
+ "selected_residual_scale_counts": {
239
+ "0.35": 288,
240
+ "0.45": 237,
241
+ "0.4": 50
242
+ },
243
+ "policy_rollout_success_rate": 0.35478260869565215,
244
+ "policy_rollout_progress": 0.5724870564092113,
245
+ "oracle_success_rate": 0.8695652173913043,
246
+ "action_mse_to_best": 0.5053197572339813,
247
+ "best_policy_val": {
248
+ "bc_loss": 0.31955528176493114,
249
+ "field_effect_loss": 0.010269693254182736,
250
+ "field_potential_loss": 0.30970407525698346,
251
+ "field_preference_loss": 0.5266202224625481,
252
+ "lattice_edges": 3833.3333333333335,
253
+ "progress_mae": 0.20275857630703184,
254
+ "proposal_loss": 0.0,
255
+ "rank_acc": 0.8124439285861121,
256
+ "rank_loss": 0.5266130566596985,
257
+ "regret_mae": 0.40278742214043933,
258
+ "success_accuracy": 0.8677195310592651,
259
+ "total_loss": 0.2706700298521254,
260
+ "transport_field_candidates": 0.0,
261
+ "transport_field_edges": 0.0,
262
+ "transport_field_loss": 0.0,
263
+ "transport_field_potential_loss": 0.0,
264
+ "transport_field_preference_loss": 0.0,
265
+ "transport_field_rank_acc": 0.0
266
+ },
267
+ "per_task": {
268
+ "LiftPegUpright-v1": {
269
+ "action_mse_to_best": 0.4109904438677549,
270
+ "expert_success_rate": 0.8584070796460177,
271
+ "num_groups": 113,
272
+ "oracle_success_rate": 0.9380530973451328,
273
+ "policy_expert_regret": 0.968556373520235,
274
+ "policy_oracle_regret": 1.047717105498356,
275
+ "policy_rollout_progress": 0.6366955052434871,
276
+ "policy_rollout_success_rate": 0.24778761061946902,
277
+ "restore_max_error": 2.384185791015625e-07
278
+ },
279
+ "PickCube-v1": {
280
+ "action_mse_to_best": 0.4489570604637265,
281
+ "expert_success_rate": 0.9402173913043478,
282
+ "num_groups": 184,
283
+ "oracle_success_rate": 0.9456521739130435,
284
+ "policy_expert_regret": 1.089701304283074,
285
+ "policy_oracle_regret": 1.0972382092504236,
286
+ "policy_rollout_progress": 0.5621739127167055,
287
+ "policy_rollout_success_rate": 0.2717391304347826,
288
+ "restore_max_error": 2.384185791015625e-07
289
+ },
290
+ "PullCube-v1": {
291
+ "action_mse_to_best": 0.6314871114139494,
292
+ "expert_success_rate": 0.25,
293
+ "num_groups": 76,
294
+ "oracle_success_rate": 0.40789473684210525,
295
+ "policy_expert_regret": 0.2960355247885577,
296
+ "policy_oracle_regret": 0.5195493033817528,
297
+ "policy_rollout_progress": 0.29346823355330054,
298
+ "policy_rollout_success_rate": 0.18421052631578946,
299
+ "restore_max_error": 2.384185791015625e-07
300
+ },
301
+ "PushCube-v1": {
302
+ "action_mse_to_best": 0.42584498293764955,
303
+ "expert_success_rate": 0.8198198198198198,
304
+ "num_groups": 111,
305
+ "oracle_success_rate": 1.0,
306
+ "policy_expert_regret": 0.28062590352586797,
307
+ "policy_oracle_regret": 0.31423999986670037,
308
+ "policy_rollout_progress": 0.8479221622954618,
309
+ "policy_rollout_success_rate": 0.8378378378378378,
310
+ "restore_max_error": 2.3795291781425476e-07
311
+ },
312
+ "StackCube-v1": {
313
+ "action_mse_to_best": 0.7279893137760215,
314
+ "expert_success_rate": 0.6923076923076923,
315
+ "num_groups": 91,
316
+ "oracle_success_rate": 0.8571428571428571,
317
+ "policy_expert_regret": 1.0655161363410426,
318
+ "policy_oracle_regret": 1.212466462464123,
319
+ "policy_rollout_progress": 0.41066505096770906,
320
+ "policy_rollout_success_rate": 0.2087912087912088,
321
+ "restore_max_error": 1.9371509552001953e-07
322
+ }
323
+ }
324
+ },
325
+ {
326
+ "seed": 2,
327
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/transport_field_reground_fieldonly_k6clean_dropnoopwg_b12_v1/seed_2/policy_rollout_fieldonly_k6clean_dropnoopwg_b12_v1_besttransport_margin0p00_k6_srcscore001_metric_taskrel.json",
328
+ "num_groups": 575,
329
+ "selection_mode": "retrieval_residual",
330
+ "num_candidates": 48,
331
+ "candidate_sigma": 0.0,
332
+ "proposal_types": [],
333
+ "proposal_lattice_types": [],
334
+ "selection_margin": 0.0,
335
+ "prepend_policy_candidate": false,
336
+ "field_optim_steps": 0,
337
+ "field_optim_step_size": 0.0,
338
+ "field_optim_trust_radius": 0.0,
339
+ "field_optim_l2_penalty": 0.0,
340
+ "retrieval_neighbors": 6,
341
+ "retrieval_metric": "task_relative",
342
+ "retrieval_type_min_success": 0.0,
343
+ "retrieval_type_success_bonus_scale": 0.0,
344
+ "retrieval_residual_consensus_penalty_scale": 0.0,
345
+ "retrieval_residual_min_source_progress": 0.0,
346
+ "retrieval_residual_min_source_advantage": -1000000000.0,
347
+ "retrieval_residual_source_progress_bonus_scale": 0.0,
348
+ "retrieval_residual_source_score_bonus_scale": 0.01,
349
+ "retrieval_residual_source_advantage_bonus_scale": 0.0,
350
+ "retrieval_residual_composite_l2_penalty_scale": 0.0,
351
+ "retrieval_residual_scale": 1.0,
352
+ "retrieval_residual_scales": [
353
+ 0.35,
354
+ 0.4,
355
+ 0.45
356
+ ],
357
+ "retrieval_residual_anchor": "expert",
358
+ "retrieval_residual_direction": "candidate_minus_anchor",
359
+ "retrieval_residual_reduce": "compose_mean_by_type",
360
+ "retrieval_residual_challenger_types": [],
361
+ "retrieval_residual_challenger_scales": [],
362
+ "retrieval_residual_challenger_margin": 0.0,
363
+ "retrieval_residual_challenger_type_margins": {},
364
+ "retrieval_residual_challenger_tasks": [],
365
+ "retrieval_residual_challenger_type_tasks": {},
366
+ "lattice_exclude_types": [
367
+ "residual_random_negative",
368
+ "residual_wrong_direction",
369
+ "residual_near_miss+residual_no_op",
370
+ "residual_no_op+residual_wrong_gripper"
371
+ ],
372
+ "lattice_exclude_type_tasks": {},
373
+ "candidate_type_bonuses": {},
374
+ "candidate_type_bonuses_by_task": {},
375
+ "candidate_type_bonus_components": false,
376
+ "candidate_oracle_rollouts": 0,
377
+ "candidate_oracle_unique_tolerance": 1e-06,
378
+ "candidate_oracle_success_rate": null,
379
+ "candidate_oracle_progress": null,
380
+ "candidate_oracle_score": null,
381
+ "candidate_oracle_regret": null,
382
+ "candidate_oracle_score_gain_over_selected": null,
383
+ "candidate_oracle_unique_count": null,
384
+ "candidate_oracle_improvement_rate": null,
385
+ "candidate_oracle_selected_branch_success_rate": null,
386
+ "candidate_oracle_selected_branch_progress": null,
387
+ "candidate_oracle_type_counts": {},
388
+ "candidate_oracle_best_branch_rank": null,
389
+ "candidate_oracle_best_branch_rank_counts": {},
390
+ "candidate_oracle_branch_success_rates": [],
391
+ "candidate_oracle_branch_progress": [],
392
+ "candidate_oracle_branch_score_gains_over_selected": [],
393
+ "selected_residual_scale_counts": {
394
+ "0.35": 296,
395
+ "0.45": 204,
396
+ "0.4": 75
397
+ },
398
+ "policy_rollout_success_rate": 0.38782608695652177,
399
+ "policy_rollout_progress": 0.5961601831666802,
400
+ "oracle_success_rate": 0.8765217391304347,
401
+ "action_mse_to_best": 0.533600711837087,
402
+ "best_policy_val": {
403
+ "bc_loss": 0.2974168997671869,
404
+ "field_effect_loss": 0.009809228720971279,
405
+ "field_potential_loss": 0.29470594475666684,
406
+ "field_preference_loss": 0.5080636441707611,
407
+ "lattice_edges": 3833.3333333333335,
408
+ "progress_mae": 0.1944621221886741,
409
+ "proposal_loss": 0.0,
410
+ "rank_acc": 0.8252745270729065,
411
+ "rank_loss": 0.5080037944846683,
412
+ "regret_mae": 0.3989163488149643,
413
+ "success_accuracy": 0.870691223276986,
414
+ "total_loss": 0.2595750474267536,
415
+ "transport_field_candidates": 0.0,
416
+ "transport_field_edges": 0.0,
417
+ "transport_field_loss": 0.0,
418
+ "transport_field_potential_loss": 0.0,
419
+ "transport_field_preference_loss": 0.0,
420
+ "transport_field_rank_acc": 0.0
421
+ },
422
+ "per_task": {
423
+ "LiftPegUpright-v1": {
424
+ "action_mse_to_best": 0.4306636511561616,
425
+ "expert_success_rate": 0.8229166666666666,
426
+ "num_groups": 96,
427
+ "oracle_success_rate": 0.9270833333333334,
428
+ "policy_expert_regret": 0.8179064658470452,
429
+ "policy_oracle_regret": 0.9109260463155806,
430
+ "policy_rollout_progress": 0.6580991436106464,
431
+ "policy_rollout_success_rate": 0.3645833333333333,
432
+ "restore_max_error": 1.955777406692505e-07
433
+ },
434
+ "PickCube-v1": {
435
+ "action_mse_to_best": 0.5025136812467768,
436
+ "expert_success_rate": 0.9444444444444444,
437
+ "num_groups": 198,
438
+ "oracle_success_rate": 0.9595959595959596,
439
+ "policy_expert_regret": 0.9676363274097594,
440
+ "policy_oracle_regret": 0.9747197779384676,
441
+ "policy_rollout_progress": 0.6337008711722012,
442
+ "policy_rollout_success_rate": 0.35353535353535354,
443
+ "restore_max_error": 2.384185791015625e-07
444
+ },
445
+ "PullCube-v1": {
446
+ "action_mse_to_best": 0.6643689109219445,
447
+ "expert_success_rate": 0.24444444444444444,
448
+ "num_groups": 90,
449
+ "oracle_success_rate": 0.4666666666666667,
450
+ "policy_expert_regret": 0.31186816998653943,
451
+ "policy_oracle_regret": 0.4993719957975878,
452
+ "policy_rollout_progress": 0.34372065288739073,
453
+ "policy_rollout_success_rate": 0.23333333333333334,
454
+ "restore_max_error": 2.384185791015625e-07
455
+ },
456
+ "PushCube-v1": {
457
+ "action_mse_to_best": 0.5080124131671273,
458
+ "expert_success_rate": 0.8514851485148515,
459
+ "num_groups": 101,
460
+ "oracle_success_rate": 1.0,
461
+ "policy_expert_regret": 0.30721950855585606,
462
+ "policy_oracle_regret": 0.3799329262853849,
463
+ "policy_rollout_progress": 0.8180868756948131,
464
+ "policy_rollout_success_rate": 0.801980198019802,
465
+ "restore_max_error": 2.384185791015625e-07
466
+ },
467
+ "StackCube-v1": {
468
+ "action_mse_to_best": 0.6097392688401871,
469
+ "expert_success_rate": 0.7666666666666667,
470
+ "num_groups": 90,
471
+ "oracle_success_rate": 0.9111111111111111,
472
+ "policy_expert_regret": 1.134208650721444,
473
+ "policy_oracle_regret": 1.2575619707504908,
474
+ "policy_rollout_progress": 0.4508909093009101,
475
+ "policy_rollout_success_rate": 0.17777777777777778,
476
+ "restore_max_error": 2.2351741790771484e-07
477
+ }
478
+ }
479
+ }
480
+ ]
481
+ }
workspace/results/h16_transport_field_reground_fieldonly_k6clean_dropnoopwg_b12_v1_besttransport_margin0p00_k6_srcscore001_metric_taskrel_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: `transport_field_reground_fieldonly_k6clean_dropnoopwg_b12_v1`
5
+ Result file: `policy_rollout_fieldonly_k6clean_dropnoopwg_b12_v1_besttransport_margin0p00_k6_srcscore001_metric_taskrel.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: 37.22% +/- 1.66%
11
+ Gain vs h=16 rank checkpoint: +7.48%
12
+ Mean progress: 58.49%
13
+ Mean action MSE to best: 0.512
14
+
15
+ | seed | mode | k | policy cand | retrieval K | retrieval metric | residual anchor | residual direction | residual reduce | min type success | type success bonus | consensus penalty | min source progress | source progress bonus | source score bonus | residual scale | residual scales | margin | sigma | opt steps | trust | success | progress | oracle | candidate oracle | oracle gain | action MSE |
16
+ |---:|---|---:|---|---:|---|---|---|---|---:|---:|---:|---:|---:|---:|---:|---|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|
17
+ | 0 | retrieval_residual | 48 | no | 6 | task_relative | expert | candidate_minus_anchor | compose_mean_by_type | 0.00 | 0.000 | 0.000 | 0.00 | 0.000 | 0.010 | 1.00 | 0.35,0.40,0.45 | 0.000 | 0.00 | 0 | 0.00 | 37.39% | 58.60% | 85.74% | n/a | n/a | 0.496 |
18
+ | 1 | retrieval_residual | 48 | no | 6 | task_relative | expert | candidate_minus_anchor | compose_mean_by_type | 0.00 | 0.000 | 0.000 | 0.00 | 0.000 | 0.010 | 1.00 | 0.35,0.40,0.45 | 0.000 | 0.00 | 0 | 0.00 | 35.48% | 57.25% | 86.96% | n/a | n/a | 0.505 |
19
+ | 2 | retrieval_residual | 48 | no | 6 | task_relative | expert | candidate_minus_anchor | compose_mean_by_type | 0.00 | 0.000 | 0.000 | 0.00 | 0.000 | 0.010 | 1.00 | 0.35,0.40,0.45 | 0.000 | 0.00 | 0 | 0.00 | 38.78% | 59.62% | 87.65% | n/a | n/a | 0.534 |
workspace/results/h16_transport_field_reground_fieldonly_k6clean_dropnoopwg_b12_v1_besttransport_margin0p00_k6_srcscore001_metric_taskrelz_summary.json ADDED
@@ -0,0 +1,481 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "run_root": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs",
3
+ "objective": "transport_field_reground_fieldonly_k6clean_dropnoopwg_b12_v1",
4
+ "out_name": "policy_rollout_fieldonly_k6clean_dropnoopwg_b12_v1_besttransport_margin0p00_k6_srcscore001_metric_taskrelz.json",
5
+ "num_completed": 3,
6
+ "baseline_h4_policy_success": 0.2967,
7
+ "baseline_h16_rank_checkpoint_success": 0.29739130434782607,
8
+ "mean_success": 0.3698550724637681,
9
+ "std_success": 0.02015691437763908,
10
+ "mean_progress": 0.5848812929805928,
11
+ "mean_action_mse_to_best": 0.5108807214875908,
12
+ "gain_vs_h4": 0.07315507246376807,
13
+ "gain_vs_h16_rank_checkpoint": 0.07246376811594202,
14
+ "rows": [
15
+ {
16
+ "seed": 0,
17
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/transport_field_reground_fieldonly_k6clean_dropnoopwg_b12_v1/seed_0/policy_rollout_fieldonly_k6clean_dropnoopwg_b12_v1_besttransport_margin0p00_k6_srcscore001_metric_taskrelz.json",
18
+ "num_groups": 575,
19
+ "selection_mode": "retrieval_residual",
20
+ "num_candidates": 48,
21
+ "candidate_sigma": 0.0,
22
+ "proposal_types": [],
23
+ "proposal_lattice_types": [],
24
+ "selection_margin": 0.0,
25
+ "prepend_policy_candidate": false,
26
+ "field_optim_steps": 0,
27
+ "field_optim_step_size": 0.0,
28
+ "field_optim_trust_radius": 0.0,
29
+ "field_optim_l2_penalty": 0.0,
30
+ "retrieval_neighbors": 6,
31
+ "retrieval_metric": "task_relative_zscore",
32
+ "retrieval_type_min_success": 0.0,
33
+ "retrieval_type_success_bonus_scale": 0.0,
34
+ "retrieval_residual_consensus_penalty_scale": 0.0,
35
+ "retrieval_residual_min_source_progress": 0.0,
36
+ "retrieval_residual_min_source_advantage": -1000000000.0,
37
+ "retrieval_residual_source_progress_bonus_scale": 0.0,
38
+ "retrieval_residual_source_score_bonus_scale": 0.01,
39
+ "retrieval_residual_source_advantage_bonus_scale": 0.0,
40
+ "retrieval_residual_composite_l2_penalty_scale": 0.0,
41
+ "retrieval_residual_scale": 1.0,
42
+ "retrieval_residual_scales": [
43
+ 0.35,
44
+ 0.4,
45
+ 0.45
46
+ ],
47
+ "retrieval_residual_anchor": "expert",
48
+ "retrieval_residual_direction": "candidate_minus_anchor",
49
+ "retrieval_residual_reduce": "compose_mean_by_type",
50
+ "retrieval_residual_challenger_types": [],
51
+ "retrieval_residual_challenger_scales": [],
52
+ "retrieval_residual_challenger_margin": 0.0,
53
+ "retrieval_residual_challenger_type_margins": {},
54
+ "retrieval_residual_challenger_tasks": [],
55
+ "retrieval_residual_challenger_type_tasks": {},
56
+ "lattice_exclude_types": [
57
+ "residual_random_negative",
58
+ "residual_wrong_direction",
59
+ "residual_near_miss+residual_no_op",
60
+ "residual_no_op+residual_wrong_gripper"
61
+ ],
62
+ "lattice_exclude_type_tasks": {},
63
+ "candidate_type_bonuses": {},
64
+ "candidate_type_bonuses_by_task": {},
65
+ "candidate_type_bonus_components": false,
66
+ "candidate_oracle_rollouts": 0,
67
+ "candidate_oracle_unique_tolerance": 1e-06,
68
+ "candidate_oracle_success_rate": null,
69
+ "candidate_oracle_progress": null,
70
+ "candidate_oracle_score": null,
71
+ "candidate_oracle_regret": null,
72
+ "candidate_oracle_score_gain_over_selected": null,
73
+ "candidate_oracle_unique_count": null,
74
+ "candidate_oracle_improvement_rate": null,
75
+ "candidate_oracle_selected_branch_success_rate": null,
76
+ "candidate_oracle_selected_branch_progress": null,
77
+ "candidate_oracle_type_counts": {},
78
+ "candidate_oracle_best_branch_rank": null,
79
+ "candidate_oracle_best_branch_rank_counts": {},
80
+ "candidate_oracle_branch_success_rates": [],
81
+ "candidate_oracle_branch_progress": [],
82
+ "candidate_oracle_branch_score_gains_over_selected": [],
83
+ "selected_residual_scale_counts": {
84
+ "0.35": 329,
85
+ "0.45": 186,
86
+ "0.4": 60
87
+ },
88
+ "policy_rollout_success_rate": 0.3565217391304348,
89
+ "policy_rollout_progress": 0.5738849096746771,
90
+ "oracle_success_rate": 0.8573913043478261,
91
+ "action_mse_to_best": 0.4926190784978478,
92
+ "best_policy_val": {
93
+ "bc_loss": 0.3445456624031067,
94
+ "field_effect_loss": 0.009637858977334367,
95
+ "field_potential_loss": 0.28480341533819836,
96
+ "field_preference_loss": 0.5244891477955712,
97
+ "lattice_edges": 3833.3333333333335,
98
+ "progress_mae": 0.19576937291357252,
99
+ "proposal_loss": 0.0,
100
+ "rank_acc": 0.8192069795396593,
101
+ "rank_loss": 0.524484740363227,
102
+ "regret_mae": 0.42191632754272884,
103
+ "success_accuracy": 0.8693191342883639,
104
+ "total_loss": 0.261704473859734,
105
+ "transport_field_candidates": 0.0,
106
+ "transport_field_edges": 0.0,
107
+ "transport_field_loss": 0.0,
108
+ "transport_field_potential_loss": 0.0,
109
+ "transport_field_preference_loss": 0.0,
110
+ "transport_field_rank_acc": 0.0
111
+ },
112
+ "per_task": {
113
+ "LiftPegUpright-v1": {
114
+ "action_mse_to_best": 0.340019617008871,
115
+ "expert_success_rate": 0.8865979381443299,
116
+ "num_groups": 97,
117
+ "oracle_success_rate": 0.9278350515463918,
118
+ "policy_expert_regret": 1.0782840038390504,
119
+ "policy_oracle_regret": 1.1046994815474933,
120
+ "policy_rollout_progress": 0.5715128897698885,
121
+ "policy_rollout_success_rate": 0.2268041237113402,
122
+ "restore_max_error": 2.384185791015625e-07
123
+ },
124
+ "PickCube-v1": {
125
+ "action_mse_to_best": 0.4601181654008822,
126
+ "expert_success_rate": 0.9375,
127
+ "num_groups": 208,
128
+ "oracle_success_rate": 0.9471153846153846,
129
+ "policy_expert_regret": 0.9763227530952113,
130
+ "policy_oracle_regret": 0.9897456205217168,
131
+ "policy_rollout_progress": 0.6105592753469514,
132
+ "policy_rollout_success_rate": 0.3317307692307692,
133
+ "restore_max_error": 2.384185791015625e-07
134
+ },
135
+ "PullCube-v1": {
136
+ "action_mse_to_best": 0.6306563861571349,
137
+ "expert_success_rate": 0.19480519480519481,
138
+ "num_groups": 77,
139
+ "oracle_success_rate": 0.36363636363636365,
140
+ "policy_expert_regret": 0.2693978397817466,
141
+ "policy_oracle_regret": 0.48080125034903176,
142
+ "policy_rollout_progress": 0.3176867231350942,
143
+ "policy_rollout_success_rate": 0.2077922077922078,
144
+ "restore_max_error": 2.3096799850463867e-07
145
+ },
146
+ "PushCube-v1": {
147
+ "action_mse_to_best": 0.4702776832941999,
148
+ "expert_success_rate": 0.8617021276595744,
149
+ "num_groups": 94,
150
+ "oracle_success_rate": 0.9893617021276596,
151
+ "policy_expert_regret": 0.436042198792417,
152
+ "policy_oracle_regret": 0.49452738939447605,
153
+ "policy_rollout_progress": 0.7607917595416942,
154
+ "policy_rollout_success_rate": 0.7446808510638298,
155
+ "restore_max_error": 2.384185791015625e-07
156
+ },
157
+ "StackCube-v1": {
158
+ "action_mse_to_best": 0.6242711608006497,
159
+ "expert_success_rate": 0.6767676767676768,
160
+ "num_groups": 99,
161
+ "oracle_success_rate": 0.8585858585858586,
162
+ "policy_expert_regret": 0.8760737712034071,
163
+ "policy_oracle_regret": 1.040449914456618,
164
+ "policy_rollout_progress": 0.5209539434825531,
165
+ "policy_rollout_success_rate": 0.2828282828282828,
166
+ "restore_max_error": 2.384185791015625e-07
167
+ }
168
+ }
169
+ },
170
+ {
171
+ "seed": 1,
172
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/transport_field_reground_fieldonly_k6clean_dropnoopwg_b12_v1/seed_1/policy_rollout_fieldonly_k6clean_dropnoopwg_b12_v1_besttransport_margin0p00_k6_srcscore001_metric_taskrelz.json",
173
+ "num_groups": 575,
174
+ "selection_mode": "retrieval_residual",
175
+ "num_candidates": 48,
176
+ "candidate_sigma": 0.0,
177
+ "proposal_types": [],
178
+ "proposal_lattice_types": [],
179
+ "selection_margin": 0.0,
180
+ "prepend_policy_candidate": false,
181
+ "field_optim_steps": 0,
182
+ "field_optim_step_size": 0.0,
183
+ "field_optim_trust_radius": 0.0,
184
+ "field_optim_l2_penalty": 0.0,
185
+ "retrieval_neighbors": 6,
186
+ "retrieval_metric": "task_relative_zscore",
187
+ "retrieval_type_min_success": 0.0,
188
+ "retrieval_type_success_bonus_scale": 0.0,
189
+ "retrieval_residual_consensus_penalty_scale": 0.0,
190
+ "retrieval_residual_min_source_progress": 0.0,
191
+ "retrieval_residual_min_source_advantage": -1000000000.0,
192
+ "retrieval_residual_source_progress_bonus_scale": 0.0,
193
+ "retrieval_residual_source_score_bonus_scale": 0.01,
194
+ "retrieval_residual_source_advantage_bonus_scale": 0.0,
195
+ "retrieval_residual_composite_l2_penalty_scale": 0.0,
196
+ "retrieval_residual_scale": 1.0,
197
+ "retrieval_residual_scales": [
198
+ 0.35,
199
+ 0.4,
200
+ 0.45
201
+ ],
202
+ "retrieval_residual_anchor": "expert",
203
+ "retrieval_residual_direction": "candidate_minus_anchor",
204
+ "retrieval_residual_reduce": "compose_mean_by_type",
205
+ "retrieval_residual_challenger_types": [],
206
+ "retrieval_residual_challenger_scales": [],
207
+ "retrieval_residual_challenger_margin": 0.0,
208
+ "retrieval_residual_challenger_type_margins": {},
209
+ "retrieval_residual_challenger_tasks": [],
210
+ "retrieval_residual_challenger_type_tasks": {},
211
+ "lattice_exclude_types": [
212
+ "residual_random_negative",
213
+ "residual_wrong_direction",
214
+ "residual_near_miss+residual_no_op",
215
+ "residual_no_op+residual_wrong_gripper"
216
+ ],
217
+ "lattice_exclude_type_tasks": {},
218
+ "candidate_type_bonuses": {},
219
+ "candidate_type_bonuses_by_task": {},
220
+ "candidate_type_bonus_components": false,
221
+ "candidate_oracle_rollouts": 0,
222
+ "candidate_oracle_unique_tolerance": 1e-06,
223
+ "candidate_oracle_success_rate": null,
224
+ "candidate_oracle_progress": null,
225
+ "candidate_oracle_score": null,
226
+ "candidate_oracle_regret": null,
227
+ "candidate_oracle_score_gain_over_selected": null,
228
+ "candidate_oracle_unique_count": null,
229
+ "candidate_oracle_improvement_rate": null,
230
+ "candidate_oracle_selected_branch_success_rate": null,
231
+ "candidate_oracle_selected_branch_progress": null,
232
+ "candidate_oracle_type_counts": {},
233
+ "candidate_oracle_best_branch_rank": null,
234
+ "candidate_oracle_best_branch_rank_counts": {},
235
+ "candidate_oracle_branch_success_rates": [],
236
+ "candidate_oracle_branch_progress": [],
237
+ "candidate_oracle_branch_score_gains_over_selected": [],
238
+ "selected_residual_scale_counts": {
239
+ "0.35": 289,
240
+ "0.45": 236,
241
+ "0.4": 50
242
+ },
243
+ "policy_rollout_success_rate": 0.36,
244
+ "policy_rollout_progress": 0.5792882018633511,
245
+ "oracle_success_rate": 0.8695652173913043,
246
+ "action_mse_to_best": 0.5073619219069572,
247
+ "best_policy_val": {
248
+ "bc_loss": 0.31955528176493114,
249
+ "field_effect_loss": 0.010269693254182736,
250
+ "field_potential_loss": 0.30970407525698346,
251
+ "field_preference_loss": 0.5266202224625481,
252
+ "lattice_edges": 3833.3333333333335,
253
+ "progress_mae": 0.20275857630703184,
254
+ "proposal_loss": 0.0,
255
+ "rank_acc": 0.8124439285861121,
256
+ "rank_loss": 0.5266130566596985,
257
+ "regret_mae": 0.40278742214043933,
258
+ "success_accuracy": 0.8677195310592651,
259
+ "total_loss": 0.2706700298521254,
260
+ "transport_field_candidates": 0.0,
261
+ "transport_field_edges": 0.0,
262
+ "transport_field_loss": 0.0,
263
+ "transport_field_potential_loss": 0.0,
264
+ "transport_field_preference_loss": 0.0,
265
+ "transport_field_rank_acc": 0.0
266
+ },
267
+ "per_task": {
268
+ "LiftPegUpright-v1": {
269
+ "action_mse_to_best": 0.4108842911602583,
270
+ "expert_success_rate": 0.8584070796460177,
271
+ "num_groups": 113,
272
+ "oracle_success_rate": 0.9380530973451328,
273
+ "policy_expert_regret": 0.9685817305493144,
274
+ "policy_oracle_regret": 1.0477621331151608,
275
+ "policy_rollout_progress": 0.6366504776266824,
276
+ "policy_rollout_success_rate": 0.24778761061946902,
277
+ "restore_max_error": 2.384185791015625e-07
278
+ },
279
+ "PickCube-v1": {
280
+ "action_mse_to_best": 0.4507029051689998,
281
+ "expert_success_rate": 0.9402173913043478,
282
+ "num_groups": 184,
283
+ "oracle_success_rate": 0.9456521739130435,
284
+ "policy_expert_regret": 1.0667215416754556,
285
+ "policy_oracle_regret": 1.0742584466428051,
286
+ "policy_rollout_progress": 0.5764021896655954,
287
+ "policy_rollout_success_rate": 0.28804347826086957,
288
+ "restore_max_error": 2.384185791015625e-07
289
+ },
290
+ "PullCube-v1": {
291
+ "action_mse_to_best": 0.6240704349781337,
292
+ "expert_success_rate": 0.25,
293
+ "num_groups": 76,
294
+ "oracle_success_rate": 0.40789473684210525,
295
+ "policy_expert_regret": 0.27225225989853863,
296
+ "policy_oracle_regret": 0.5179581383586322,
297
+ "policy_rollout_progress": 0.2944453668192421,
298
+ "policy_rollout_success_rate": 0.18421052631578946,
299
+ "restore_max_error": 2.384185791015625e-07
300
+ },
301
+ "PushCube-v1": {
302
+ "action_mse_to_best": 0.42836992536579166,
303
+ "expert_success_rate": 0.8198198198198198,
304
+ "num_groups": 111,
305
+ "oracle_success_rate": 1.0,
306
+ "policy_expert_regret": 0.2802314224006893,
307
+ "policy_oracle_regret": 0.31377608443165683,
308
+ "policy_rollout_progress": 0.8483860777305053,
309
+ "policy_rollout_success_rate": 0.8378378378378378,
310
+ "restore_max_error": 2.3795291781425476e-07
311
+ },
312
+ "StackCube-v1": {
313
+ "action_mse_to_best": 0.7406091304434524,
314
+ "expert_success_rate": 0.6923076923076923,
315
+ "num_groups": 91,
316
+ "oracle_success_rate": 0.8571428571428571,
317
+ "policy_expert_regret": 1.077667276119138,
318
+ "policy_oracle_regret": 1.2121162426013212,
319
+ "policy_rollout_progress": 0.4235440296131176,
320
+ "policy_rollout_success_rate": 0.2087912087912088,
321
+ "restore_max_error": 1.9371509552001953e-07
322
+ }
323
+ }
324
+ },
325
+ {
326
+ "seed": 2,
327
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/transport_field_reground_fieldonly_k6clean_dropnoopwg_b12_v1/seed_2/policy_rollout_fieldonly_k6clean_dropnoopwg_b12_v1_besttransport_margin0p00_k6_srcscore001_metric_taskrelz.json",
328
+ "num_groups": 575,
329
+ "selection_mode": "retrieval_residual",
330
+ "num_candidates": 48,
331
+ "candidate_sigma": 0.0,
332
+ "proposal_types": [],
333
+ "proposal_lattice_types": [],
334
+ "selection_margin": 0.0,
335
+ "prepend_policy_candidate": false,
336
+ "field_optim_steps": 0,
337
+ "field_optim_step_size": 0.0,
338
+ "field_optim_trust_radius": 0.0,
339
+ "field_optim_l2_penalty": 0.0,
340
+ "retrieval_neighbors": 6,
341
+ "retrieval_metric": "task_relative_zscore",
342
+ "retrieval_type_min_success": 0.0,
343
+ "retrieval_type_success_bonus_scale": 0.0,
344
+ "retrieval_residual_consensus_penalty_scale": 0.0,
345
+ "retrieval_residual_min_source_progress": 0.0,
346
+ "retrieval_residual_min_source_advantage": -1000000000.0,
347
+ "retrieval_residual_source_progress_bonus_scale": 0.0,
348
+ "retrieval_residual_source_score_bonus_scale": 0.01,
349
+ "retrieval_residual_source_advantage_bonus_scale": 0.0,
350
+ "retrieval_residual_composite_l2_penalty_scale": 0.0,
351
+ "retrieval_residual_scale": 1.0,
352
+ "retrieval_residual_scales": [
353
+ 0.35,
354
+ 0.4,
355
+ 0.45
356
+ ],
357
+ "retrieval_residual_anchor": "expert",
358
+ "retrieval_residual_direction": "candidate_minus_anchor",
359
+ "retrieval_residual_reduce": "compose_mean_by_type",
360
+ "retrieval_residual_challenger_types": [],
361
+ "retrieval_residual_challenger_scales": [],
362
+ "retrieval_residual_challenger_margin": 0.0,
363
+ "retrieval_residual_challenger_type_margins": {},
364
+ "retrieval_residual_challenger_tasks": [],
365
+ "retrieval_residual_challenger_type_tasks": {},
366
+ "lattice_exclude_types": [
367
+ "residual_random_negative",
368
+ "residual_wrong_direction",
369
+ "residual_near_miss+residual_no_op",
370
+ "residual_no_op+residual_wrong_gripper"
371
+ ],
372
+ "lattice_exclude_type_tasks": {},
373
+ "candidate_type_bonuses": {},
374
+ "candidate_type_bonuses_by_task": {},
375
+ "candidate_type_bonus_components": false,
376
+ "candidate_oracle_rollouts": 0,
377
+ "candidate_oracle_unique_tolerance": 1e-06,
378
+ "candidate_oracle_success_rate": null,
379
+ "candidate_oracle_progress": null,
380
+ "candidate_oracle_score": null,
381
+ "candidate_oracle_regret": null,
382
+ "candidate_oracle_score_gain_over_selected": null,
383
+ "candidate_oracle_unique_count": null,
384
+ "candidate_oracle_improvement_rate": null,
385
+ "candidate_oracle_selected_branch_success_rate": null,
386
+ "candidate_oracle_selected_branch_progress": null,
387
+ "candidate_oracle_type_counts": {},
388
+ "candidate_oracle_best_branch_rank": null,
389
+ "candidate_oracle_best_branch_rank_counts": {},
390
+ "candidate_oracle_branch_success_rates": [],
391
+ "candidate_oracle_branch_progress": [],
392
+ "candidate_oracle_branch_score_gains_over_selected": [],
393
+ "selected_residual_scale_counts": {
394
+ "0.35": 300,
395
+ "0.45": 195,
396
+ "0.4": 80
397
+ },
398
+ "policy_rollout_success_rate": 0.39304347826086955,
399
+ "policy_rollout_progress": 0.6014707674037503,
400
+ "oracle_success_rate": 0.8765217391304347,
401
+ "action_mse_to_best": 0.5326611640579675,
402
+ "best_policy_val": {
403
+ "bc_loss": 0.2974168997671869,
404
+ "field_effect_loss": 0.009809228720971279,
405
+ "field_potential_loss": 0.29470594475666684,
406
+ "field_preference_loss": 0.5080636441707611,
407
+ "lattice_edges": 3833.3333333333335,
408
+ "progress_mae": 0.1944621221886741,
409
+ "proposal_loss": 0.0,
410
+ "rank_acc": 0.8252745270729065,
411
+ "rank_loss": 0.5080037944846683,
412
+ "regret_mae": 0.3989163488149643,
413
+ "success_accuracy": 0.870691223276986,
414
+ "total_loss": 0.2595750474267536,
415
+ "transport_field_candidates": 0.0,
416
+ "transport_field_edges": 0.0,
417
+ "transport_field_loss": 0.0,
418
+ "transport_field_potential_loss": 0.0,
419
+ "transport_field_preference_loss": 0.0,
420
+ "transport_field_rank_acc": 0.0
421
+ },
422
+ "per_task": {
423
+ "LiftPegUpright-v1": {
424
+ "action_mse_to_best": 0.4301719583551555,
425
+ "expert_success_rate": 0.8229166666666666,
426
+ "num_groups": 96,
427
+ "oracle_success_rate": 0.9270833333333334,
428
+ "policy_expert_regret": 0.8189836051315069,
429
+ "policy_oracle_regret": 0.9120031856000423,
430
+ "policy_rollout_progress": 0.6570220043261846,
431
+ "policy_rollout_success_rate": 0.3645833333333333,
432
+ "restore_max_error": 1.955777406692505e-07
433
+ },
434
+ "PickCube-v1": {
435
+ "action_mse_to_best": 0.49772106083768486,
436
+ "expert_success_rate": 0.9444444444444444,
437
+ "num_groups": 198,
438
+ "oracle_success_rate": 0.9595959595959596,
439
+ "policy_expert_regret": 0.9751971047653845,
440
+ "policy_oracle_regret": 0.9822805552940929,
441
+ "policy_rollout_progress": 0.636241103917586,
442
+ "policy_rollout_success_rate": 0.3434343434343434,
443
+ "restore_max_error": 2.384185791015625e-07
444
+ },
445
+ "PullCube-v1": {
446
+ "action_mse_to_best": 0.6605500661664538,
447
+ "expert_success_rate": 0.24444444444444444,
448
+ "num_groups": 90,
449
+ "oracle_success_rate": 0.4666666666666667,
450
+ "policy_expert_regret": 0.3101457569644683,
451
+ "policy_oracle_regret": 0.49626951777479716,
452
+ "policy_rollout_progress": 0.35522911733844215,
453
+ "policy_rollout_success_rate": 0.24444444444444444,
454
+ "restore_max_error": 2.384185791015625e-07
455
+ },
456
+ "PushCube-v1": {
457
+ "action_mse_to_best": 0.5114603226285169,
458
+ "expert_success_rate": 0.8514851485148515,
459
+ "num_groups": 101,
460
+ "oracle_success_rate": 1.0,
461
+ "policy_expert_regret": 0.26736475010909655,
462
+ "policy_oracle_regret": 0.3036904891233633,
463
+ "policy_rollout_progress": 0.8547253524607951,
464
+ "policy_rollout_success_rate": 0.8415841584158416,
465
+ "restore_max_error": 2.384185791015625e-07
466
+ },
467
+ "StackCube-v1": {
468
+ "action_mse_to_best": 0.6147543638323744,
469
+ "expert_success_rate": 0.7666666666666667,
470
+ "num_groups": 90,
471
+ "oracle_success_rate": 0.9111111111111111,
472
+ "policy_expert_regret": 1.1399508054057756,
473
+ "policy_oracle_regret": 1.2764820527699259,
474
+ "policy_rollout_progress": 0.427755101190673,
475
+ "policy_rollout_success_rate": 0.17777777777777778,
476
+ "restore_max_error": 2.2351741790771484e-07
477
+ }
478
+ }
479
+ }
480
+ ]
481
+ }
workspace/results/h16_transport_field_reground_fieldonly_k6clean_dropnoopwg_b12_v1_besttransport_margin0p00_k6_srcscore001_metric_taskrelz_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: `transport_field_reground_fieldonly_k6clean_dropnoopwg_b12_v1`
5
+ Result file: `policy_rollout_fieldonly_k6clean_dropnoopwg_b12_v1_besttransport_margin0p00_k6_srcscore001_metric_taskrelz.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: 36.99% +/- 2.02%
11
+ Gain vs h=16 rank checkpoint: +7.25%
12
+ Mean progress: 58.49%
13
+ Mean action MSE to best: 0.511
14
+
15
+ | seed | mode | k | policy cand | retrieval K | retrieval metric | residual anchor | residual direction | residual reduce | min type success | type success bonus | consensus penalty | min source progress | source progress bonus | source score bonus | residual scale | residual scales | margin | sigma | opt steps | trust | success | progress | oracle | candidate oracle | oracle gain | action MSE |
16
+ |---:|---|---:|---|---:|---|---|---|---|---:|---:|---:|---:|---:|---:|---:|---|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|
17
+ | 0 | retrieval_residual | 48 | no | 6 | task_relative_zscore | expert | candidate_minus_anchor | compose_mean_by_type | 0.00 | 0.000 | 0.000 | 0.00 | 0.000 | 0.010 | 1.00 | 0.35,0.40,0.45 | 0.000 | 0.00 | 0 | 0.00 | 35.65% | 57.39% | 85.74% | n/a | n/a | 0.493 |
18
+ | 1 | retrieval_residual | 48 | no | 6 | task_relative_zscore | expert | candidate_minus_anchor | compose_mean_by_type | 0.00 | 0.000 | 0.000 | 0.00 | 0.000 | 0.010 | 1.00 | 0.35,0.40,0.45 | 0.000 | 0.00 | 0 | 0.00 | 36.00% | 57.93% | 86.96% | n/a | n/a | 0.507 |
19
+ | 2 | retrieval_residual | 48 | no | 6 | task_relative_zscore | expert | candidate_minus_anchor | compose_mean_by_type | 0.00 | 0.000 | 0.000 | 0.00 | 0.000 | 0.010 | 1.00 | 0.35,0.40,0.45 | 0.000 | 0.00 | 0 | 0.00 | 39.30% | 60.15% | 87.65% | n/a | n/a | 0.533 |
workspace/results/h16_transport_field_reground_fieldonly_k6clean_dropnoopwg_b12_v1_besttransport_margin0p00_k6_srcscore001_metric_zscore_summary.json ADDED
@@ -0,0 +1,481 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "run_root": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs",
3
+ "objective": "transport_field_reground_fieldonly_k6clean_dropnoopwg_b12_v1",
4
+ "out_name": "policy_rollout_fieldonly_k6clean_dropnoopwg_b12_v1_besttransport_margin0p00_k6_srcscore001_metric_zscore.json",
5
+ "num_completed": 3,
6
+ "baseline_h4_policy_success": 0.2967,
7
+ "baseline_h16_rank_checkpoint_success": 0.29739130434782607,
8
+ "mean_success": 0.3785507246376812,
9
+ "std_success": 0.012215250727945201,
10
+ "mean_progress": 0.5908983214655419,
11
+ "mean_action_mse_to_best": 0.5220838308536813,
12
+ "gain_vs_h4": 0.08185072463768117,
13
+ "gain_vs_h16_rank_checkpoint": 0.08115942028985512,
14
+ "rows": [
15
+ {
16
+ "seed": 0,
17
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/transport_field_reground_fieldonly_k6clean_dropnoopwg_b12_v1/seed_0/policy_rollout_fieldonly_k6clean_dropnoopwg_b12_v1_besttransport_margin0p00_k6_srcscore001_metric_zscore.json",
18
+ "num_groups": 575,
19
+ "selection_mode": "retrieval_residual",
20
+ "num_candidates": 48,
21
+ "candidate_sigma": 0.0,
22
+ "proposal_types": [],
23
+ "proposal_lattice_types": [],
24
+ "selection_margin": 0.0,
25
+ "prepend_policy_candidate": false,
26
+ "field_optim_steps": 0,
27
+ "field_optim_step_size": 0.0,
28
+ "field_optim_trust_radius": 0.0,
29
+ "field_optim_l2_penalty": 0.0,
30
+ "retrieval_neighbors": 6,
31
+ "retrieval_metric": "zscore",
32
+ "retrieval_type_min_success": 0.0,
33
+ "retrieval_type_success_bonus_scale": 0.0,
34
+ "retrieval_residual_consensus_penalty_scale": 0.0,
35
+ "retrieval_residual_min_source_progress": 0.0,
36
+ "retrieval_residual_min_source_advantage": -1000000000.0,
37
+ "retrieval_residual_source_progress_bonus_scale": 0.0,
38
+ "retrieval_residual_source_score_bonus_scale": 0.01,
39
+ "retrieval_residual_source_advantage_bonus_scale": 0.0,
40
+ "retrieval_residual_composite_l2_penalty_scale": 0.0,
41
+ "retrieval_residual_scale": 1.0,
42
+ "retrieval_residual_scales": [
43
+ 0.35,
44
+ 0.4,
45
+ 0.45
46
+ ],
47
+ "retrieval_residual_anchor": "expert",
48
+ "retrieval_residual_direction": "candidate_minus_anchor",
49
+ "retrieval_residual_reduce": "compose_mean_by_type",
50
+ "retrieval_residual_challenger_types": [],
51
+ "retrieval_residual_challenger_scales": [],
52
+ "retrieval_residual_challenger_margin": 0.0,
53
+ "retrieval_residual_challenger_type_margins": {},
54
+ "retrieval_residual_challenger_tasks": [],
55
+ "retrieval_residual_challenger_type_tasks": {},
56
+ "lattice_exclude_types": [
57
+ "residual_random_negative",
58
+ "residual_wrong_direction",
59
+ "residual_near_miss+residual_no_op",
60
+ "residual_no_op+residual_wrong_gripper"
61
+ ],
62
+ "lattice_exclude_type_tasks": {},
63
+ "candidate_type_bonuses": {},
64
+ "candidate_type_bonuses_by_task": {},
65
+ "candidate_type_bonus_components": false,
66
+ "candidate_oracle_rollouts": 0,
67
+ "candidate_oracle_unique_tolerance": 1e-06,
68
+ "candidate_oracle_success_rate": null,
69
+ "candidate_oracle_progress": null,
70
+ "candidate_oracle_score": null,
71
+ "candidate_oracle_regret": null,
72
+ "candidate_oracle_score_gain_over_selected": null,
73
+ "candidate_oracle_unique_count": null,
74
+ "candidate_oracle_improvement_rate": null,
75
+ "candidate_oracle_selected_branch_success_rate": null,
76
+ "candidate_oracle_selected_branch_progress": null,
77
+ "candidate_oracle_type_counts": {},
78
+ "candidate_oracle_best_branch_rank": null,
79
+ "candidate_oracle_best_branch_rank_counts": {},
80
+ "candidate_oracle_branch_success_rates": [],
81
+ "candidate_oracle_branch_progress": [],
82
+ "candidate_oracle_branch_score_gains_over_selected": [],
83
+ "selected_residual_scale_counts": {
84
+ "0.35": 319,
85
+ "0.45": 167,
86
+ "0.4": 89
87
+ },
88
+ "policy_rollout_success_rate": 0.3773913043478261,
89
+ "policy_rollout_progress": 0.5823645833001027,
90
+ "oracle_success_rate": 0.8573913043478261,
91
+ "action_mse_to_best": 0.5014481430354973,
92
+ "best_policy_val": {
93
+ "bc_loss": 0.3445456624031067,
94
+ "field_effect_loss": 0.009637858977334367,
95
+ "field_potential_loss": 0.28480341533819836,
96
+ "field_preference_loss": 0.5244891477955712,
97
+ "lattice_edges": 3833.3333333333335,
98
+ "progress_mae": 0.19576937291357252,
99
+ "proposal_loss": 0.0,
100
+ "rank_acc": 0.8192069795396593,
101
+ "rank_loss": 0.524484740363227,
102
+ "regret_mae": 0.42191632754272884,
103
+ "success_accuracy": 0.8693191342883639,
104
+ "total_loss": 0.261704473859734,
105
+ "transport_field_candidates": 0.0,
106
+ "transport_field_edges": 0.0,
107
+ "transport_field_loss": 0.0,
108
+ "transport_field_potential_loss": 0.0,
109
+ "transport_field_preference_loss": 0.0,
110
+ "transport_field_rank_acc": 0.0
111
+ },
112
+ "per_task": {
113
+ "LiftPegUpright-v1": {
114
+ "action_mse_to_best": 0.342714158501284,
115
+ "expert_success_rate": 0.8865979381443299,
116
+ "num_groups": 97,
117
+ "oracle_success_rate": 0.9278350515463918,
118
+ "policy_expert_regret": 1.0783122260238707,
119
+ "policy_oracle_regret": 1.1047277037323135,
120
+ "policy_rollout_progress": 0.5714846675850681,
121
+ "policy_rollout_success_rate": 0.2268041237113402,
122
+ "restore_max_error": 2.384185791015625e-07
123
+ },
124
+ "PickCube-v1": {
125
+ "action_mse_to_best": 0.47303284792444455,
126
+ "expert_success_rate": 0.9375,
127
+ "num_groups": 208,
128
+ "oracle_success_rate": 0.9471153846153846,
129
+ "policy_expert_regret": 0.9706395817831015,
130
+ "policy_oracle_regret": 0.9706608165109244,
131
+ "policy_rollout_progress": 0.6104133101269746,
132
+ "policy_rollout_success_rate": 0.35096153846153844,
133
+ "restore_max_error": 2.384185791015625e-07
134
+ },
135
+ "PullCube-v1": {
136
+ "action_mse_to_best": 0.6115741692967229,
137
+ "expert_success_rate": 0.19480519480519481,
138
+ "num_groups": 77,
139
+ "oracle_success_rate": 0.36363636363636365,
140
+ "policy_expert_regret": 0.21880711770205238,
141
+ "policy_oracle_regret": 0.36467767267657275,
142
+ "policy_rollout_progress": 0.37832084527638343,
143
+ "policy_rollout_success_rate": 0.2857142857142857,
144
+ "restore_max_error": 2.3096799850463867e-07
145
+ },
146
+ "PushCube-v1": {
147
+ "action_mse_to_best": 0.5068147898988521,
148
+ "expert_success_rate": 0.8617021276595744,
149
+ "num_groups": 94,
150
+ "oracle_success_rate": 0.9893617021276596,
151
+ "policy_expert_regret": 0.41722126780672275,
152
+ "policy_oracle_regret": 0.4757848258348221,
153
+ "policy_rollout_progress": 0.7599067114135052,
154
+ "policy_rollout_success_rate": 0.7446808510638298,
155
+ "restore_max_error": 2.384185791015625e-07
156
+ },
157
+ "StackCube-v1": {
158
+ "action_mse_to_best": 0.6259270223854768,
159
+ "expert_success_rate": 0.6767676767676768,
160
+ "num_groups": 99,
161
+ "oracle_success_rate": 0.8585858585858586,
162
+ "policy_expert_regret": 0.8642329713611892,
163
+ "policy_oracle_regret": 1.0054746368015655,
164
+ "policy_rollout_progress": 0.5242193755176332,
165
+ "policy_rollout_success_rate": 0.30303030303030304,
166
+ "restore_max_error": 2.384185791015625e-07
167
+ }
168
+ }
169
+ },
170
+ {
171
+ "seed": 1,
172
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/transport_field_reground_fieldonly_k6clean_dropnoopwg_b12_v1/seed_1/policy_rollout_fieldonly_k6clean_dropnoopwg_b12_v1_besttransport_margin0p00_k6_srcscore001_metric_zscore.json",
173
+ "num_groups": 575,
174
+ "selection_mode": "retrieval_residual",
175
+ "num_candidates": 48,
176
+ "candidate_sigma": 0.0,
177
+ "proposal_types": [],
178
+ "proposal_lattice_types": [],
179
+ "selection_margin": 0.0,
180
+ "prepend_policy_candidate": false,
181
+ "field_optim_steps": 0,
182
+ "field_optim_step_size": 0.0,
183
+ "field_optim_trust_radius": 0.0,
184
+ "field_optim_l2_penalty": 0.0,
185
+ "retrieval_neighbors": 6,
186
+ "retrieval_metric": "zscore",
187
+ "retrieval_type_min_success": 0.0,
188
+ "retrieval_type_success_bonus_scale": 0.0,
189
+ "retrieval_residual_consensus_penalty_scale": 0.0,
190
+ "retrieval_residual_min_source_progress": 0.0,
191
+ "retrieval_residual_min_source_advantage": -1000000000.0,
192
+ "retrieval_residual_source_progress_bonus_scale": 0.0,
193
+ "retrieval_residual_source_score_bonus_scale": 0.01,
194
+ "retrieval_residual_source_advantage_bonus_scale": 0.0,
195
+ "retrieval_residual_composite_l2_penalty_scale": 0.0,
196
+ "retrieval_residual_scale": 1.0,
197
+ "retrieval_residual_scales": [
198
+ 0.35,
199
+ 0.4,
200
+ 0.45
201
+ ],
202
+ "retrieval_residual_anchor": "expert",
203
+ "retrieval_residual_direction": "candidate_minus_anchor",
204
+ "retrieval_residual_reduce": "compose_mean_by_type",
205
+ "retrieval_residual_challenger_types": [],
206
+ "retrieval_residual_challenger_scales": [],
207
+ "retrieval_residual_challenger_margin": 0.0,
208
+ "retrieval_residual_challenger_type_margins": {},
209
+ "retrieval_residual_challenger_tasks": [],
210
+ "retrieval_residual_challenger_type_tasks": {},
211
+ "lattice_exclude_types": [
212
+ "residual_random_negative",
213
+ "residual_wrong_direction",
214
+ "residual_near_miss+residual_no_op",
215
+ "residual_no_op+residual_wrong_gripper"
216
+ ],
217
+ "lattice_exclude_type_tasks": {},
218
+ "candidate_type_bonuses": {},
219
+ "candidate_type_bonuses_by_task": {},
220
+ "candidate_type_bonus_components": false,
221
+ "candidate_oracle_rollouts": 0,
222
+ "candidate_oracle_unique_tolerance": 1e-06,
223
+ "candidate_oracle_success_rate": null,
224
+ "candidate_oracle_progress": null,
225
+ "candidate_oracle_score": null,
226
+ "candidate_oracle_regret": null,
227
+ "candidate_oracle_score_gain_over_selected": null,
228
+ "candidate_oracle_unique_count": null,
229
+ "candidate_oracle_improvement_rate": null,
230
+ "candidate_oracle_selected_branch_success_rate": null,
231
+ "candidate_oracle_selected_branch_progress": null,
232
+ "candidate_oracle_type_counts": {},
233
+ "candidate_oracle_best_branch_rank": null,
234
+ "candidate_oracle_best_branch_rank_counts": {},
235
+ "candidate_oracle_branch_success_rates": [],
236
+ "candidate_oracle_branch_progress": [],
237
+ "candidate_oracle_branch_score_gains_over_selected": [],
238
+ "selected_residual_scale_counts": {
239
+ "0.35": 285,
240
+ "0.4": 61,
241
+ "0.45": 229
242
+ },
243
+ "policy_rollout_success_rate": 0.36695652173913046,
244
+ "policy_rollout_progress": 0.5842054872760428,
245
+ "oracle_success_rate": 0.8695652173913043,
246
+ "action_mse_to_best": 0.5208085850974464,
247
+ "best_policy_val": {
248
+ "bc_loss": 0.31955528176493114,
249
+ "field_effect_loss": 0.010269693254182736,
250
+ "field_potential_loss": 0.30970407525698346,
251
+ "field_preference_loss": 0.5266202224625481,
252
+ "lattice_edges": 3833.3333333333335,
253
+ "progress_mae": 0.20275857630703184,
254
+ "proposal_loss": 0.0,
255
+ "rank_acc": 0.8124439285861121,
256
+ "rank_loss": 0.5266130566596985,
257
+ "regret_mae": 0.40278742214043933,
258
+ "success_accuracy": 0.8677195310592651,
259
+ "total_loss": 0.2706700298521254,
260
+ "transport_field_candidates": 0.0,
261
+ "transport_field_edges": 0.0,
262
+ "transport_field_loss": 0.0,
263
+ "transport_field_potential_loss": 0.0,
264
+ "transport_field_preference_loss": 0.0,
265
+ "transport_field_rank_acc": 0.0
266
+ },
267
+ "per_task": {
268
+ "LiftPegUpright-v1": {
269
+ "action_mse_to_best": 0.42033933778499477,
270
+ "expert_success_rate": 0.8584070796460177,
271
+ "num_groups": 113,
272
+ "oracle_success_rate": 0.9380530973451328,
273
+ "policy_expert_regret": 0.9413453508003623,
274
+ "policy_oracle_regret": 1.0207694108243537,
275
+ "policy_rollout_progress": 0.6435028927779831,
276
+ "policy_rollout_success_rate": 0.25663716814159293,
277
+ "restore_max_error": 2.384185791015625e-07
278
+ },
279
+ "PickCube-v1": {
280
+ "action_mse_to_best": 0.4699172475982619,
281
+ "expert_success_rate": 0.9402173913043478,
282
+ "num_groups": 184,
283
+ "oracle_success_rate": 0.9456521739130435,
284
+ "policy_expert_regret": 1.0882801643785331,
285
+ "policy_oracle_regret": 1.0883041688534634,
286
+ "policy_rollout_progress": 0.5623564674549372,
287
+ "policy_rollout_success_rate": 0.28804347826086957,
288
+ "restore_max_error": 2.384185791015625e-07
289
+ },
290
+ "PullCube-v1": {
291
+ "action_mse_to_best": 0.6565579777877582,
292
+ "expert_success_rate": 0.25,
293
+ "num_groups": 76,
294
+ "oracle_success_rate": 0.40789473684210525,
295
+ "policy_expert_regret": 0.24084414361966897,
296
+ "policy_oracle_regret": 0.41145320909823857,
297
+ "policy_rollout_progress": 0.3492067722246774,
298
+ "policy_rollout_success_rate": 0.23684210526315788,
299
+ "restore_max_error": 2.384185791015625e-07
300
+ },
301
+ "PushCube-v1": {
302
+ "action_mse_to_best": 0.43571353549355857,
303
+ "expert_success_rate": 0.8198198198198198,
304
+ "num_groups": 111,
305
+ "oracle_success_rate": 1.0,
306
+ "policy_expert_regret": 0.27903562220367223,
307
+ "policy_oracle_regret": 0.31258028423463974,
308
+ "policy_rollout_progress": 0.8495818779275224,
309
+ "policy_rollout_success_rate": 0.8378378378378378,
310
+ "restore_max_error": 2.3795291781425476e-07
311
+ },
312
+ "StackCube-v1": {
313
+ "action_mse_to_best": 0.7388924060614555,
314
+ "expert_success_rate": 0.6923076923076923,
315
+ "num_groups": 91,
316
+ "oracle_success_rate": 0.8571428571428571,
317
+ "policy_expert_regret": 1.0595285558438563,
318
+ "policy_oracle_regret": 1.2067449872310345,
319
+ "policy_rollout_progress": 0.42731247416564394,
320
+ "policy_rollout_success_rate": 0.1978021978021978,
321
+ "restore_max_error": 1.9371509552001953e-07
322
+ }
323
+ }
324
+ },
325
+ {
326
+ "seed": 2,
327
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/transport_field_reground_fieldonly_k6clean_dropnoopwg_b12_v1/seed_2/policy_rollout_fieldonly_k6clean_dropnoopwg_b12_v1_besttransport_margin0p00_k6_srcscore001_metric_zscore.json",
328
+ "num_groups": 575,
329
+ "selection_mode": "retrieval_residual",
330
+ "num_candidates": 48,
331
+ "candidate_sigma": 0.0,
332
+ "proposal_types": [],
333
+ "proposal_lattice_types": [],
334
+ "selection_margin": 0.0,
335
+ "prepend_policy_candidate": false,
336
+ "field_optim_steps": 0,
337
+ "field_optim_step_size": 0.0,
338
+ "field_optim_trust_radius": 0.0,
339
+ "field_optim_l2_penalty": 0.0,
340
+ "retrieval_neighbors": 6,
341
+ "retrieval_metric": "zscore",
342
+ "retrieval_type_min_success": 0.0,
343
+ "retrieval_type_success_bonus_scale": 0.0,
344
+ "retrieval_residual_consensus_penalty_scale": 0.0,
345
+ "retrieval_residual_min_source_progress": 0.0,
346
+ "retrieval_residual_min_source_advantage": -1000000000.0,
347
+ "retrieval_residual_source_progress_bonus_scale": 0.0,
348
+ "retrieval_residual_source_score_bonus_scale": 0.01,
349
+ "retrieval_residual_source_advantage_bonus_scale": 0.0,
350
+ "retrieval_residual_composite_l2_penalty_scale": 0.0,
351
+ "retrieval_residual_scale": 1.0,
352
+ "retrieval_residual_scales": [
353
+ 0.35,
354
+ 0.4,
355
+ 0.45
356
+ ],
357
+ "retrieval_residual_anchor": "expert",
358
+ "retrieval_residual_direction": "candidate_minus_anchor",
359
+ "retrieval_residual_reduce": "compose_mean_by_type",
360
+ "retrieval_residual_challenger_types": [],
361
+ "retrieval_residual_challenger_scales": [],
362
+ "retrieval_residual_challenger_margin": 0.0,
363
+ "retrieval_residual_challenger_type_margins": {},
364
+ "retrieval_residual_challenger_tasks": [],
365
+ "retrieval_residual_challenger_type_tasks": {},
366
+ "lattice_exclude_types": [
367
+ "residual_random_negative",
368
+ "residual_wrong_direction",
369
+ "residual_near_miss+residual_no_op",
370
+ "residual_no_op+residual_wrong_gripper"
371
+ ],
372
+ "lattice_exclude_type_tasks": {},
373
+ "candidate_type_bonuses": {},
374
+ "candidate_type_bonuses_by_task": {},
375
+ "candidate_type_bonus_components": false,
376
+ "candidate_oracle_rollouts": 0,
377
+ "candidate_oracle_unique_tolerance": 1e-06,
378
+ "candidate_oracle_success_rate": null,
379
+ "candidate_oracle_progress": null,
380
+ "candidate_oracle_score": null,
381
+ "candidate_oracle_regret": null,
382
+ "candidate_oracle_score_gain_over_selected": null,
383
+ "candidate_oracle_unique_count": null,
384
+ "candidate_oracle_improvement_rate": null,
385
+ "candidate_oracle_selected_branch_success_rate": null,
386
+ "candidate_oracle_selected_branch_progress": null,
387
+ "candidate_oracle_type_counts": {},
388
+ "candidate_oracle_best_branch_rank": null,
389
+ "candidate_oracle_best_branch_rank_counts": {},
390
+ "candidate_oracle_branch_success_rates": [],
391
+ "candidate_oracle_branch_progress": [],
392
+ "candidate_oracle_branch_score_gains_over_selected": [],
393
+ "selected_residual_scale_counts": {
394
+ "0.35": 317,
395
+ "0.45": 191,
396
+ "0.4": 67
397
+ },
398
+ "policy_rollout_success_rate": 0.391304347826087,
399
+ "policy_rollout_progress": 0.6061248938204802,
400
+ "oracle_success_rate": 0.8765217391304347,
401
+ "action_mse_to_best": 0.5439947644280999,
402
+ "best_policy_val": {
403
+ "bc_loss": 0.2974168997671869,
404
+ "field_effect_loss": 0.009809228720971279,
405
+ "field_potential_loss": 0.29470594475666684,
406
+ "field_preference_loss": 0.5080636441707611,
407
+ "lattice_edges": 3833.3333333333335,
408
+ "progress_mae": 0.1944621221886741,
409
+ "proposal_loss": 0.0,
410
+ "rank_acc": 0.8252745270729065,
411
+ "rank_loss": 0.5080037944846683,
412
+ "regret_mae": 0.3989163488149643,
413
+ "success_accuracy": 0.870691223276986,
414
+ "total_loss": 0.2595750474267536,
415
+ "transport_field_candidates": 0.0,
416
+ "transport_field_edges": 0.0,
417
+ "transport_field_loss": 0.0,
418
+ "transport_field_potential_loss": 0.0,
419
+ "transport_field_preference_loss": 0.0,
420
+ "transport_field_rank_acc": 0.0
421
+ },
422
+ "per_task": {
423
+ "LiftPegUpright-v1": {
424
+ "action_mse_to_best": 0.44746410054115887,
425
+ "expert_success_rate": 0.8229166666666666,
426
+ "num_groups": 96,
427
+ "oracle_success_rate": 0.9270833333333334,
428
+ "policy_expert_regret": 0.8183209470783671,
429
+ "policy_oracle_regret": 0.9113291455432773,
430
+ "policy_rollout_progress": 0.6576960443829497,
431
+ "policy_rollout_success_rate": 0.3645833333333333,
432
+ "restore_max_error": 1.955777406692505e-07
433
+ },
434
+ "PickCube-v1": {
435
+ "action_mse_to_best": 0.49729833526141715,
436
+ "expert_success_rate": 0.9444444444444444,
437
+ "num_groups": 198,
438
+ "oracle_success_rate": 0.9595959595959596,
439
+ "policy_expert_regret": 0.9677926309764234,
440
+ "policy_oracle_regret": 0.9748760815051318,
441
+ "policy_rollout_progress": 0.6385950726560421,
442
+ "policy_rollout_success_rate": 0.3484848484848485,
443
+ "restore_max_error": 2.384185791015625e-07
444
+ },
445
+ "PullCube-v1": {
446
+ "action_mse_to_best": 0.6874315675761965,
447
+ "expert_success_rate": 0.24444444444444444,
448
+ "num_groups": 90,
449
+ "oracle_success_rate": 0.4666666666666667,
450
+ "policy_expert_regret": 0.34411713575116465,
451
+ "policy_oracle_regret": 0.4879133890796867,
452
+ "policy_rollout_progress": 0.36900863573989934,
453
+ "policy_rollout_success_rate": 0.25555555555555554,
454
+ "restore_max_error": 2.384185791015625e-07
455
+ },
456
+ "PushCube-v1": {
457
+ "action_mse_to_best": 0.5177686682136933,
458
+ "expert_success_rate": 0.8514851485148515,
459
+ "num_groups": 101,
460
+ "oracle_success_rate": 1.0,
461
+ "policy_expert_regret": 0.3584785184057632,
462
+ "policy_oracle_regret": 0.3584785184057632,
463
+ "policy_rollout_progress": 0.8296402934754249,
464
+ "policy_rollout_success_rate": 0.8118811881188119,
465
+ "restore_max_error": 2.384185791015625e-07
466
+ },
467
+ "StackCube-v1": {
468
+ "action_mse_to_best": 0.6356876549000542,
469
+ "expert_success_rate": 0.7666666666666667,
470
+ "num_groups": 90,
471
+ "oracle_success_rate": 0.9111111111111111,
472
+ "policy_expert_regret": 1.114454825884766,
473
+ "policy_oracle_regret": 1.2382735712660684,
474
+ "policy_rollout_progress": 0.4659635826945305,
475
+ "policy_rollout_success_rate": 0.17777777777777778,
476
+ "restore_max_error": 2.2351741790771484e-07
477
+ }
478
+ }
479
+ }
480
+ ]
481
+ }
workspace/results/h16_transport_field_reground_fieldonly_k6clean_dropnoopwg_b12_v1_besttransport_margin0p00_k6_srcscore001_metric_zscore_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: `transport_field_reground_fieldonly_k6clean_dropnoopwg_b12_v1`
5
+ Result file: `policy_rollout_fieldonly_k6clean_dropnoopwg_b12_v1_besttransport_margin0p00_k6_srcscore001_metric_zscore.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: 37.86% +/- 1.22%
11
+ Gain vs h=16 rank checkpoint: +8.12%
12
+ Mean progress: 59.09%
13
+ Mean action MSE to best: 0.522
14
+
15
+ | seed | mode | k | policy cand | retrieval K | retrieval metric | residual anchor | residual direction | residual reduce | min type success | type success bonus | consensus penalty | min source progress | source progress bonus | source score bonus | residual scale | residual scales | margin | sigma | opt steps | trust | success | progress | oracle | candidate oracle | oracle gain | action MSE |
16
+ |---:|---|---:|---|---:|---|---|---|---|---:|---:|---:|---:|---:|---:|---:|---|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|
17
+ | 0 | retrieval_residual | 48 | no | 6 | zscore | expert | candidate_minus_anchor | compose_mean_by_type | 0.00 | 0.000 | 0.000 | 0.00 | 0.000 | 0.010 | 1.00 | 0.35,0.40,0.45 | 0.000 | 0.00 | 0 | 0.00 | 37.74% | 58.24% | 85.74% | n/a | n/a | 0.501 |
18
+ | 1 | retrieval_residual | 48 | no | 6 | zscore | expert | candidate_minus_anchor | compose_mean_by_type | 0.00 | 0.000 | 0.000 | 0.00 | 0.000 | 0.010 | 1.00 | 0.35,0.40,0.45 | 0.000 | 0.00 | 0 | 0.00 | 36.70% | 58.42% | 86.96% | n/a | n/a | 0.521 |
19
+ | 2 | retrieval_residual | 48 | no | 6 | zscore | expert | candidate_minus_anchor | compose_mean_by_type | 0.00 | 0.000 | 0.000 | 0.00 | 0.000 | 0.010 | 1.00 | 0.35,0.40,0.45 | 0.000 | 0.00 | 0 | 0.00 | 39.13% | 60.61% | 87.65% | n/a | n/a | 0.544 |