anhtld commited on
Commit
9ccb91b
·
verified ·
1 Parent(s): 67742bb

Auto-sync: 2026-06-30 04:47:32 (part 2)

Browse files
results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmchallenger0p005_summary.json ADDED
@@ -0,0 +1,445 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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_nmchallenger0p005.json",
5
+ "num_completed": 3,
6
+ "baseline_h4_policy_success": 0.2967,
7
+ "baseline_h16_rank_checkpoint_success": 0.29739130434782607,
8
+ "mean_success": 0.36,
9
+ "std_success": 0.014235396124995555,
10
+ "mean_progress": 0.5734310006964412,
11
+ "mean_action_mse_to_best": 0.40723808390803745,
12
+ "gain_vs_h4": 0.06329999999999997,
13
+ "gain_vs_h16_rank_checkpoint": 0.06260869565217392,
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_nmchallenger0p005.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
+ "retrieval_residual_challenger_types": [
49
+ "residual_near_miss"
50
+ ],
51
+ "retrieval_residual_challenger_margin": 0.005,
52
+ "lattice_exclude_types": [
53
+ "residual_random_negative",
54
+ "residual_wrong_direction",
55
+ "residual_near_miss+residual_no_op"
56
+ ],
57
+ "candidate_type_bonuses": {
58
+ "residual_no_op": 0.03
59
+ },
60
+ "candidate_type_bonus_components": false,
61
+ "candidate_oracle_rollouts": 0,
62
+ "candidate_oracle_unique_tolerance": 1e-06,
63
+ "candidate_oracle_success_rate": null,
64
+ "candidate_oracle_progress": null,
65
+ "candidate_oracle_score": null,
66
+ "candidate_oracle_regret": null,
67
+ "candidate_oracle_score_gain_over_selected": null,
68
+ "candidate_oracle_unique_count": null,
69
+ "candidate_oracle_improvement_rate": null,
70
+ "candidate_oracle_selected_branch_success_rate": null,
71
+ "candidate_oracle_selected_branch_progress": null,
72
+ "candidate_oracle_type_counts": {},
73
+ "candidate_oracle_best_branch_rank": null,
74
+ "candidate_oracle_best_branch_rank_counts": {},
75
+ "candidate_oracle_branch_success_rates": [],
76
+ "candidate_oracle_branch_progress": [],
77
+ "candidate_oracle_branch_score_gains_over_selected": [],
78
+ "selected_residual_scale_counts": {
79
+ "0.35": 514,
80
+ "0.45": 56,
81
+ "0.4": 5
82
+ },
83
+ "policy_rollout_success_rate": 0.3565217391304348,
84
+ "policy_rollout_progress": 0.5589558385697234,
85
+ "oracle_success_rate": 0.8573913043478261,
86
+ "action_mse_to_best": 0.39503803297391404,
87
+ "best_policy_val": {
88
+ "bc_loss": 0.13721593966086706,
89
+ "field_effect_loss": 0.009290305380192068,
90
+ "field_potential_loss": 0.2666468388504452,
91
+ "field_preference_loss": 0.5130573478009965,
92
+ "lattice_edges": 3833.3333333333335,
93
+ "progress_mae": 0.1933159919248687,
94
+ "rank_acc": 0.8265031774838766,
95
+ "rank_loss": 0.5130523675017886,
96
+ "regret_mae": 0.3756548762321472,
97
+ "success_accuracy": 0.8773836526605818,
98
+ "total_loss": 1.5581054819954767
99
+ },
100
+ "per_task": {
101
+ "LiftPegUpright-v1": {
102
+ "action_mse_to_best": 0.32276735697709713,
103
+ "expert_success_rate": 0.8865979381443299,
104
+ "num_groups": 97,
105
+ "oracle_success_rate": 0.9278350515463918,
106
+ "policy_expert_regret": 1.078290365866779,
107
+ "policy_oracle_regret": 1.1046864339678557,
108
+ "policy_rollout_progress": 0.571525937349526,
109
+ "policy_rollout_success_rate": 0.2268041237113402,
110
+ "restore_max_error": 2.384185791015625e-07
111
+ },
112
+ "PickCube-v1": {
113
+ "action_mse_to_best": 0.3028235242057305,
114
+ "expert_success_rate": 0.9375,
115
+ "num_groups": 208,
116
+ "oracle_success_rate": 0.9471153846153846,
117
+ "policy_expert_regret": 0.9510513620916754,
118
+ "policy_oracle_regret": 0.9644742295181808,
119
+ "policy_rollout_progress": 0.6165998971197181,
120
+ "policy_rollout_success_rate": 0.35096153846153844,
121
+ "restore_max_error": 2.2910535335540771e-07
122
+ },
123
+ "PullCube-v1": {
124
+ "action_mse_to_best": 0.6010933086856619,
125
+ "expert_success_rate": 0.19480519480519481,
126
+ "num_groups": 77,
127
+ "oracle_success_rate": 0.36363636363636365,
128
+ "policy_expert_regret": 0.2427287653554231,
129
+ "policy_oracle_regret": 0.41602517896960794,
130
+ "policy_rollout_progress": 0.3327872604101461,
131
+ "policy_rollout_success_rate": 0.23376623376623376,
132
+ "restore_max_error": 2.3096799850463867e-07
133
+ },
134
+ "PushCube-v1": {
135
+ "action_mse_to_best": 0.3688891567289829,
136
+ "expert_success_rate": 0.8617021276595744,
137
+ "num_groups": 94,
138
+ "oracle_success_rate": 0.9893617021276596,
139
+ "policy_expert_regret": 0.5723979790159996,
140
+ "policy_oracle_regret": 0.6499486254884842,
141
+ "policy_rollout_progress": 0.6904769064264095,
142
+ "policy_rollout_success_rate": 0.6595744680851063,
143
+ "restore_max_error": 1.1920928955078125e-07
144
+ },
145
+ "StackCube-v1": {
146
+ "action_mse_to_best": 0.5241553211829277,
147
+ "expert_success_rate": 0.6767676767676768,
148
+ "num_groups": 99,
149
+ "oracle_success_rate": 0.8585858585858586,
150
+ "policy_expert_regret": 0.9052192963132955,
151
+ "policy_oracle_regret": 1.0528768245318922,
152
+ "policy_rollout_progress": 0.47655923634466496,
153
+ "policy_rollout_success_rate": 0.30303030303030304,
154
+ "restore_max_error": 2.1606683731079102e-07
155
+ }
156
+ }
157
+ },
158
+ {
159
+ "seed": 1,
160
+ "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_nmchallenger0p005.json",
161
+ "num_groups": 575,
162
+ "selection_mode": "retrieval_residual",
163
+ "num_candidates": 48,
164
+ "candidate_sigma": 0.0,
165
+ "selection_margin": 0.2,
166
+ "prepend_policy_candidate": false,
167
+ "field_optim_steps": 0,
168
+ "field_optim_step_size": 0.0,
169
+ "field_optim_trust_radius": 0.0,
170
+ "field_optim_l2_penalty": 0.0,
171
+ "retrieval_neighbors": 4,
172
+ "retrieval_metric": "raw",
173
+ "retrieval_type_min_success": 0.0,
174
+ "retrieval_type_success_bonus_scale": 0.0,
175
+ "retrieval_residual_consensus_penalty_scale": 0.0,
176
+ "retrieval_residual_min_source_progress": 0.0,
177
+ "retrieval_residual_min_source_advantage": -1000000000.0,
178
+ "retrieval_residual_source_progress_bonus_scale": 0.0,
179
+ "retrieval_residual_source_score_bonus_scale": 0.0,
180
+ "retrieval_residual_source_advantage_bonus_scale": 0.0,
181
+ "retrieval_residual_composite_l2_penalty_scale": 0.0,
182
+ "retrieval_residual_scale": 1.0,
183
+ "retrieval_residual_scales": [
184
+ 0.35,
185
+ 0.4,
186
+ 0.45
187
+ ],
188
+ "retrieval_residual_anchor": "expert",
189
+ "retrieval_residual_direction": "candidate_minus_anchor",
190
+ "retrieval_residual_reduce": "compose_mean_by_type",
191
+ "retrieval_residual_challenger_types": [
192
+ "residual_near_miss"
193
+ ],
194
+ "retrieval_residual_challenger_margin": 0.005,
195
+ "lattice_exclude_types": [
196
+ "residual_random_negative",
197
+ "residual_wrong_direction",
198
+ "residual_near_miss+residual_no_op"
199
+ ],
200
+ "candidate_type_bonuses": {
201
+ "residual_no_op": 0.03
202
+ },
203
+ "candidate_type_bonus_components": false,
204
+ "candidate_oracle_rollouts": 0,
205
+ "candidate_oracle_unique_tolerance": 1e-06,
206
+ "candidate_oracle_success_rate": null,
207
+ "candidate_oracle_progress": null,
208
+ "candidate_oracle_score": null,
209
+ "candidate_oracle_regret": null,
210
+ "candidate_oracle_score_gain_over_selected": null,
211
+ "candidate_oracle_unique_count": null,
212
+ "candidate_oracle_improvement_rate": null,
213
+ "candidate_oracle_selected_branch_success_rate": null,
214
+ "candidate_oracle_selected_branch_progress": null,
215
+ "candidate_oracle_type_counts": {},
216
+ "candidate_oracle_best_branch_rank": null,
217
+ "candidate_oracle_best_branch_rank_counts": {},
218
+ "candidate_oracle_branch_success_rates": [],
219
+ "candidate_oracle_branch_progress": [],
220
+ "candidate_oracle_branch_score_gains_over_selected": [],
221
+ "selected_residual_scale_counts": {
222
+ "0.35": 521,
223
+ "0.45": 49,
224
+ "0.4": 5
225
+ },
226
+ "policy_rollout_success_rate": 0.34782608695652173,
227
+ "policy_rollout_progress": 0.5713966150581836,
228
+ "oracle_success_rate": 0.8695652173913043,
229
+ "action_mse_to_best": 0.3989635453648541,
230
+ "best_policy_val": {
231
+ "bc_loss": 0.1315989200439718,
232
+ "field_effect_loss": 0.010110370814800262,
233
+ "field_potential_loss": 0.31546536915832096,
234
+ "field_preference_loss": 0.4960637440284093,
235
+ "lattice_edges": 3833.3333333333335,
236
+ "progress_mae": 0.19023226698239645,
237
+ "rank_acc": 0.8255469501018524,
238
+ "rank_loss": 0.4960531195004781,
239
+ "regret_mae": 0.42156532737943864,
240
+ "success_accuracy": 0.8807473679383596,
241
+ "total_loss": 1.561984618504842
242
+ },
243
+ "per_task": {
244
+ "LiftPegUpright-v1": {
245
+ "action_mse_to_best": 0.3285531557816188,
246
+ "expert_success_rate": 0.8584070796460177,
247
+ "num_groups": 113,
248
+ "oracle_success_rate": 0.9380530973451328,
249
+ "policy_expert_regret": 0.9874511470573138,
250
+ "policy_oracle_regret": 1.0666533279999169,
251
+ "policy_rollout_progress": 0.6354583977861742,
252
+ "policy_rollout_success_rate": 0.23008849557522124,
253
+ "restore_max_error": 2.384185791015625e-07
254
+ },
255
+ "PickCube-v1": {
256
+ "action_mse_to_best": 0.32256428818420874,
257
+ "expert_success_rate": 0.9402173913043478,
258
+ "num_groups": 184,
259
+ "oracle_success_rate": 0.9456521739130435,
260
+ "policy_expert_regret": 1.0746920006653375,
261
+ "policy_oracle_regret": 1.0747160051402675,
262
+ "policy_rollout_progress": 0.5705098485594372,
263
+ "policy_rollout_success_rate": 0.29347826086956524,
264
+ "restore_max_error": 2.384185791015625e-07
265
+ },
266
+ "PullCube-v1": {
267
+ "action_mse_to_best": 0.5930280594253227,
268
+ "expert_success_rate": 0.25,
269
+ "num_groups": 76,
270
+ "oracle_success_rate": 0.40789473684210525,
271
+ "policy_expert_regret": 0.2742968981742467,
272
+ "policy_oracle_regret": 0.4978899852568774,
273
+ "policy_rollout_progress": 0.30219598205171916,
274
+ "policy_rollout_success_rate": 0.19736842105263158,
275
+ "restore_max_error": 2.3096799850463867e-07
276
+ },
277
+ "PushCube-v1": {
278
+ "action_mse_to_best": 0.3595503710948669,
279
+ "expert_success_rate": 0.8198198198198198,
280
+ "num_groups": 111,
281
+ "oracle_success_rate": 1.0,
282
+ "policy_expert_regret": 0.32904101532321794,
283
+ "policy_oracle_regret": 0.34620858245604746,
284
+ "policy_rollout_progress": 0.8339715977241328,
285
+ "policy_rollout_success_rate": 0.8198198198198198,
286
+ "restore_max_error": 1.7881393432617188e-07
287
+ },
288
+ "StackCube-v1": {
289
+ "action_mse_to_best": 0.526873398326583,
290
+ "expert_success_rate": 0.6923076923076923,
291
+ "num_groups": 91,
292
+ "oracle_success_rate": 0.8571428571428571,
293
+ "policy_expert_regret": 1.1327610128855967,
294
+ "policy_oracle_regret": 1.292421804188372,
295
+ "policy_rollout_progress": 0.3981835229711218,
296
+ "policy_rollout_success_rate": 0.15384615384615385,
297
+ "restore_max_error": 1.9371509552001953e-07
298
+ }
299
+ }
300
+ },
301
+ {
302
+ "seed": 2,
303
+ "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_nmchallenger0p005.json",
304
+ "num_groups": 575,
305
+ "selection_mode": "retrieval_residual",
306
+ "num_candidates": 48,
307
+ "candidate_sigma": 0.0,
308
+ "selection_margin": 0.2,
309
+ "prepend_policy_candidate": false,
310
+ "field_optim_steps": 0,
311
+ "field_optim_step_size": 0.0,
312
+ "field_optim_trust_radius": 0.0,
313
+ "field_optim_l2_penalty": 0.0,
314
+ "retrieval_neighbors": 4,
315
+ "retrieval_metric": "raw",
316
+ "retrieval_type_min_success": 0.0,
317
+ "retrieval_type_success_bonus_scale": 0.0,
318
+ "retrieval_residual_consensus_penalty_scale": 0.0,
319
+ "retrieval_residual_min_source_progress": 0.0,
320
+ "retrieval_residual_min_source_advantage": -1000000000.0,
321
+ "retrieval_residual_source_progress_bonus_scale": 0.0,
322
+ "retrieval_residual_source_score_bonus_scale": 0.0,
323
+ "retrieval_residual_source_advantage_bonus_scale": 0.0,
324
+ "retrieval_residual_composite_l2_penalty_scale": 0.0,
325
+ "retrieval_residual_scale": 1.0,
326
+ "retrieval_residual_scales": [
327
+ 0.35,
328
+ 0.4,
329
+ 0.45
330
+ ],
331
+ "retrieval_residual_anchor": "expert",
332
+ "retrieval_residual_direction": "candidate_minus_anchor",
333
+ "retrieval_residual_reduce": "compose_mean_by_type",
334
+ "retrieval_residual_challenger_types": [
335
+ "residual_near_miss"
336
+ ],
337
+ "retrieval_residual_challenger_margin": 0.005,
338
+ "lattice_exclude_types": [
339
+ "residual_random_negative",
340
+ "residual_wrong_direction",
341
+ "residual_near_miss+residual_no_op"
342
+ ],
343
+ "candidate_type_bonuses": {
344
+ "residual_no_op": 0.03
345
+ },
346
+ "candidate_type_bonus_components": false,
347
+ "candidate_oracle_rollouts": 0,
348
+ "candidate_oracle_unique_tolerance": 1e-06,
349
+ "candidate_oracle_success_rate": null,
350
+ "candidate_oracle_progress": null,
351
+ "candidate_oracle_score": null,
352
+ "candidate_oracle_regret": null,
353
+ "candidate_oracle_score_gain_over_selected": null,
354
+ "candidate_oracle_unique_count": null,
355
+ "candidate_oracle_improvement_rate": null,
356
+ "candidate_oracle_selected_branch_success_rate": null,
357
+ "candidate_oracle_selected_branch_progress": null,
358
+ "candidate_oracle_type_counts": {},
359
+ "candidate_oracle_best_branch_rank": null,
360
+ "candidate_oracle_best_branch_rank_counts": {},
361
+ "candidate_oracle_branch_success_rates": [],
362
+ "candidate_oracle_branch_progress": [],
363
+ "candidate_oracle_branch_score_gains_over_selected": [],
364
+ "selected_residual_scale_counts": {
365
+ "0.35": 516,
366
+ "0.45": 48,
367
+ "0.4": 11
368
+ },
369
+ "policy_rollout_success_rate": 0.37565217391304345,
370
+ "policy_rollout_progress": 0.5899405484614165,
371
+ "oracle_success_rate": 0.8765217391304347,
372
+ "action_mse_to_best": 0.42771267338534413,
373
+ "best_policy_val": {
374
+ "bc_loss": 0.11367896075050037,
375
+ "field_effect_loss": 0.009670218582161598,
376
+ "field_potential_loss": 0.2641640139950646,
377
+ "field_preference_loss": 0.5130490180518892,
378
+ "lattice_edges": 3833.3333333333335,
379
+ "progress_mae": 0.2021729110015763,
380
+ "rank_acc": 0.8333857821093665,
381
+ "rank_loss": 0.5130119257503085,
382
+ "regret_mae": 0.3958987047274907,
383
+ "success_accuracy": 0.8680730561415354,
384
+ "total_loss": 1.4394984311527677
385
+ },
386
+ "per_task": {
387
+ "LiftPegUpright-v1": {
388
+ "action_mse_to_best": 0.35602021238203935,
389
+ "expert_success_rate": 0.8229166666666666,
390
+ "num_groups": 96,
391
+ "oracle_success_rate": 0.9270833333333334,
392
+ "policy_expert_regret": 0.8475987246880928,
393
+ "policy_oracle_regret": 0.9538775502393643,
394
+ "policy_rollout_progress": 0.6463976396868626,
395
+ "policy_rollout_success_rate": 0.3333333333333333,
396
+ "restore_max_error": 2.384185791015625e-07
397
+ },
398
+ "PickCube-v1": {
399
+ "action_mse_to_best": 0.33909460881517994,
400
+ "expert_success_rate": 0.9444444444444444,
401
+ "num_groups": 198,
402
+ "oracle_success_rate": 0.9595959595959596,
403
+ "policy_expert_regret": 0.9735308127910501,
404
+ "policy_oracle_regret": 0.9806142633197584,
405
+ "policy_rollout_progress": 0.6379073958919205,
406
+ "policy_rollout_success_rate": 0.3434343434343434,
407
+ "restore_max_error": 2.384185791015625e-07
408
+ },
409
+ "PullCube-v1": {
410
+ "action_mse_to_best": 0.6386238743033674,
411
+ "expert_success_rate": 0.24444444444444444,
412
+ "num_groups": 90,
413
+ "oracle_success_rate": 0.4666666666666667,
414
+ "policy_expert_regret": 0.2692028627006544,
415
+ "policy_oracle_regret": 0.4741226551019483,
416
+ "policy_rollout_progress": 0.362101345633467,
417
+ "policy_rollout_success_rate": 0.25555555555555554,
418
+ "restore_max_error": 2.384185791015625e-07
419
+ },
420
+ "PushCube-v1": {
421
+ "action_mse_to_best": 0.3970492488824495,
422
+ "expert_success_rate": 0.8514851485148515,
423
+ "num_groups": 101,
424
+ "oracle_success_rate": 1.0,
425
+ "policy_expert_regret": 0.3425388766987489,
426
+ "policy_oracle_regret": 0.39744112810285964,
427
+ "policy_rollout_progress": 0.8104796639763483,
428
+ "policy_rollout_success_rate": 0.7920792079207921,
429
+ "restore_max_error": 2.384185791015625e-07
430
+ },
431
+ "StackCube-v1": {
432
+ "action_mse_to_best": 0.5226443493117888,
433
+ "expert_success_rate": 0.7666666666666667,
434
+ "num_groups": 90,
435
+ "oracle_success_rate": 0.9111111111111111,
436
+ "policy_expert_regret": 1.1750739713509877,
437
+ "policy_oracle_regret": 1.3371944083107843,
438
+ "policy_rollout_progress": 0.4045378933350245,
439
+ "policy_rollout_success_rate": 0.14444444444444443,
440
+ "restore_max_error": 2.086162567138672e-07
441
+ }
442
+ }
443
+ }
444
+ ]
445
+ }
results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmchallenger0p005_summary.md ADDED
@@ -0,0 +1,19 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # h=16 Best-Policy Checkpoint Rollout
2
+
3
+ Run root: `/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs`
4
+ Objective: `near_miss_policy_bc5`
5
+ Result file: `policy_rollout_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmchallenger0p005.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: 36.00% +/- 1.42%
11
+ Gain vs h=16 rank checkpoint: +6.26%
12
+ Mean progress: 57.34%
13
+ Mean action MSE to best: 0.407
14
+
15
+ | seed | mode | k | policy cand | retrieval K | retrieval metric | residual anchor | residual direction | residual reduce | min type success | type success bonus | consensus penalty | min source progress | source progress bonus | source score bonus | residual scale | residual scales | margin | sigma | opt steps | trust | success | progress | oracle | candidate oracle | oracle gain | action MSE |
16
+ |---:|---|---:|---|---:|---|---|---|---|---:|---:|---:|---:|---:|---:|---:|---|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|
17
+ | 0 | retrieval_residual | 48 | no | 4 | raw | expert | candidate_minus_anchor | compose_mean_by_type + challenger residual_near_miss@0.01 | 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 | 35.65% | 55.90% | 85.74% | n/a | n/a | 0.395 |
18
+ | 1 | retrieval_residual | 48 | no | 4 | raw | expert | candidate_minus_anchor | compose_mean_by_type + challenger residual_near_miss@0.01 | 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% | 57.14% | 86.96% | n/a | n/a | 0.399 |
19
+ | 2 | retrieval_residual | 48 | no | 4 | raw | expert | candidate_minus_anchor | compose_mean_by_type + challenger residual_near_miss@0.01 | 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 | 37.57% | 58.99% | 87.65% | n/a | n/a | 0.428 |
results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmchallenger0p015_summary.json ADDED
@@ -0,0 +1,445 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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_nmchallenger0p015.json",
5
+ "num_completed": 3,
6
+ "baseline_h4_policy_success": 0.2967,
7
+ "baseline_h16_rank_checkpoint_success": 0.29739130434782607,
8
+ "mean_success": 0.36,
9
+ "std_success": 0.012541047914657353,
10
+ "mean_progress": 0.5738324999127645,
11
+ "mean_action_mse_to_best": 0.407323780797044,
12
+ "gain_vs_h4": 0.06329999999999997,
13
+ "gain_vs_h16_rank_checkpoint": 0.06260869565217392,
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_nmchallenger0p015.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
+ "retrieval_residual_challenger_types": [
49
+ "residual_near_miss"
50
+ ],
51
+ "retrieval_residual_challenger_margin": 0.015,
52
+ "lattice_exclude_types": [
53
+ "residual_random_negative",
54
+ "residual_wrong_direction",
55
+ "residual_near_miss+residual_no_op"
56
+ ],
57
+ "candidate_type_bonuses": {
58
+ "residual_no_op": 0.03
59
+ },
60
+ "candidate_type_bonus_components": false,
61
+ "candidate_oracle_rollouts": 0,
62
+ "candidate_oracle_unique_tolerance": 1e-06,
63
+ "candidate_oracle_success_rate": null,
64
+ "candidate_oracle_progress": null,
65
+ "candidate_oracle_score": null,
66
+ "candidate_oracle_regret": null,
67
+ "candidate_oracle_score_gain_over_selected": null,
68
+ "candidate_oracle_unique_count": null,
69
+ "candidate_oracle_improvement_rate": null,
70
+ "candidate_oracle_selected_branch_success_rate": null,
71
+ "candidate_oracle_selected_branch_progress": null,
72
+ "candidate_oracle_type_counts": {},
73
+ "candidate_oracle_best_branch_rank": null,
74
+ "candidate_oracle_best_branch_rank_counts": {},
75
+ "candidate_oracle_branch_success_rates": [],
76
+ "candidate_oracle_branch_progress": [],
77
+ "candidate_oracle_branch_score_gains_over_selected": [],
78
+ "selected_residual_scale_counts": {
79
+ "0.35": 518,
80
+ "0.45": 52,
81
+ "0.4": 5
82
+ },
83
+ "policy_rollout_success_rate": 0.3565217391304348,
84
+ "policy_rollout_progress": 0.559589599908289,
85
+ "oracle_success_rate": 0.8573913043478261,
86
+ "action_mse_to_best": 0.3947965855459156,
87
+ "best_policy_val": {
88
+ "bc_loss": 0.13721593966086706,
89
+ "field_effect_loss": 0.009290305380192068,
90
+ "field_potential_loss": 0.2666468388504452,
91
+ "field_preference_loss": 0.5130573478009965,
92
+ "lattice_edges": 3833.3333333333335,
93
+ "progress_mae": 0.1933159919248687,
94
+ "rank_acc": 0.8265031774838766,
95
+ "rank_loss": 0.5130523675017886,
96
+ "regret_mae": 0.3756548762321472,
97
+ "success_accuracy": 0.8773836526605818,
98
+ "total_loss": 1.5581054819954767
99
+ },
100
+ "per_task": {
101
+ "LiftPegUpright-v1": {
102
+ "action_mse_to_best": 0.32276735697709713,
103
+ "expert_success_rate": 0.8865979381443299,
104
+ "num_groups": 97,
105
+ "oracle_success_rate": 0.9278350515463918,
106
+ "policy_expert_regret": 1.078290365866779,
107
+ "policy_oracle_regret": 1.1046864339678557,
108
+ "policy_rollout_progress": 0.571525937349526,
109
+ "policy_rollout_success_rate": 0.2268041237113402,
110
+ "restore_max_error": 2.384185791015625e-07
111
+ },
112
+ "PickCube-v1": {
113
+ "action_mse_to_best": 0.30277847326718843,
114
+ "expert_success_rate": 0.9375,
115
+ "num_groups": 208,
116
+ "oracle_success_rate": 0.9471153846153846,
117
+ "policy_expert_regret": 0.9492993776220828,
118
+ "policy_oracle_regret": 0.9627222450485883,
119
+ "policy_rollout_progress": 0.6183518815893106,
120
+ "policy_rollout_success_rate": 0.35096153846153844,
121
+ "restore_max_error": 2.2910535335540771e-07
122
+ },
123
+ "PullCube-v1": {
124
+ "action_mse_to_best": 0.6010933086856619,
125
+ "expert_success_rate": 0.19480519480519481,
126
+ "num_groups": 77,
127
+ "oracle_success_rate": 0.36363636363636365,
128
+ "policy_expert_regret": 0.2427287653554231,
129
+ "policy_oracle_regret": 0.41602517896960794,
130
+ "policy_rollout_progress": 0.3327872604101461,
131
+ "policy_rollout_success_rate": 0.23376623376623376,
132
+ "restore_max_error": 2.3096799850463867e-07
133
+ },
134
+ "PushCube-v1": {
135
+ "action_mse_to_best": 0.3675119048578942,
136
+ "expert_success_rate": 0.8617021276595744,
137
+ "num_groups": 94,
138
+ "oracle_success_rate": 0.9893617021276596,
139
+ "policy_expert_regret": 0.5723979790159996,
140
+ "policy_oracle_regret": 0.6499486254884842,
141
+ "policy_rollout_progress": 0.6904769064264095,
142
+ "policy_rollout_success_rate": 0.6595744680851063,
143
+ "restore_max_error": 1.1920928955078125e-07
144
+ },
145
+ "StackCube-v1": {
146
+ "action_mse_to_best": 0.5241553211829277,
147
+ "expert_success_rate": 0.6767676767676768,
148
+ "num_groups": 99,
149
+ "oracle_success_rate": 0.8585858585858586,
150
+ "policy_expert_regret": 0.9052192963132955,
151
+ "policy_oracle_regret": 1.0528768245318922,
152
+ "policy_rollout_progress": 0.47655923634466496,
153
+ "policy_rollout_success_rate": 0.30303030303030304,
154
+ "restore_max_error": 2.1606683731079102e-07
155
+ }
156
+ }
157
+ },
158
+ {
159
+ "seed": 1,
160
+ "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_nmchallenger0p015.json",
161
+ "num_groups": 575,
162
+ "selection_mode": "retrieval_residual",
163
+ "num_candidates": 48,
164
+ "candidate_sigma": 0.0,
165
+ "selection_margin": 0.2,
166
+ "prepend_policy_candidate": false,
167
+ "field_optim_steps": 0,
168
+ "field_optim_step_size": 0.0,
169
+ "field_optim_trust_radius": 0.0,
170
+ "field_optim_l2_penalty": 0.0,
171
+ "retrieval_neighbors": 4,
172
+ "retrieval_metric": "raw",
173
+ "retrieval_type_min_success": 0.0,
174
+ "retrieval_type_success_bonus_scale": 0.0,
175
+ "retrieval_residual_consensus_penalty_scale": 0.0,
176
+ "retrieval_residual_min_source_progress": 0.0,
177
+ "retrieval_residual_min_source_advantage": -1000000000.0,
178
+ "retrieval_residual_source_progress_bonus_scale": 0.0,
179
+ "retrieval_residual_source_score_bonus_scale": 0.0,
180
+ "retrieval_residual_source_advantage_bonus_scale": 0.0,
181
+ "retrieval_residual_composite_l2_penalty_scale": 0.0,
182
+ "retrieval_residual_scale": 1.0,
183
+ "retrieval_residual_scales": [
184
+ 0.35,
185
+ 0.4,
186
+ 0.45
187
+ ],
188
+ "retrieval_residual_anchor": "expert",
189
+ "retrieval_residual_direction": "candidate_minus_anchor",
190
+ "retrieval_residual_reduce": "compose_mean_by_type",
191
+ "retrieval_residual_challenger_types": [
192
+ "residual_near_miss"
193
+ ],
194
+ "retrieval_residual_challenger_margin": 0.015,
195
+ "lattice_exclude_types": [
196
+ "residual_random_negative",
197
+ "residual_wrong_direction",
198
+ "residual_near_miss+residual_no_op"
199
+ ],
200
+ "candidate_type_bonuses": {
201
+ "residual_no_op": 0.03
202
+ },
203
+ "candidate_type_bonus_components": false,
204
+ "candidate_oracle_rollouts": 0,
205
+ "candidate_oracle_unique_tolerance": 1e-06,
206
+ "candidate_oracle_success_rate": null,
207
+ "candidate_oracle_progress": null,
208
+ "candidate_oracle_score": null,
209
+ "candidate_oracle_regret": null,
210
+ "candidate_oracle_score_gain_over_selected": null,
211
+ "candidate_oracle_unique_count": null,
212
+ "candidate_oracle_improvement_rate": null,
213
+ "candidate_oracle_selected_branch_success_rate": null,
214
+ "candidate_oracle_selected_branch_progress": null,
215
+ "candidate_oracle_type_counts": {},
216
+ "candidate_oracle_best_branch_rank": null,
217
+ "candidate_oracle_best_branch_rank_counts": {},
218
+ "candidate_oracle_branch_success_rates": [],
219
+ "candidate_oracle_branch_progress": [],
220
+ "candidate_oracle_branch_score_gains_over_selected": [],
221
+ "selected_residual_scale_counts": {
222
+ "0.35": 536,
223
+ "0.45": 36,
224
+ "0.4": 3
225
+ },
226
+ "policy_rollout_success_rate": 0.34956521739130436,
227
+ "policy_rollout_progress": 0.5728900156656037,
228
+ "oracle_success_rate": 0.8695652173913043,
229
+ "action_mse_to_best": 0.3992642119769817,
230
+ "best_policy_val": {
231
+ "bc_loss": 0.1315989200439718,
232
+ "field_effect_loss": 0.010110370814800262,
233
+ "field_potential_loss": 0.31546536915832096,
234
+ "field_preference_loss": 0.4960637440284093,
235
+ "lattice_edges": 3833.3333333333335,
236
+ "progress_mae": 0.19023226698239645,
237
+ "rank_acc": 0.8255469501018524,
238
+ "rank_loss": 0.4960531195004781,
239
+ "regret_mae": 0.42156532737943864,
240
+ "success_accuracy": 0.8807473679383596,
241
+ "total_loss": 1.561984618504842
242
+ },
243
+ "per_task": {
244
+ "LiftPegUpright-v1": {
245
+ "action_mse_to_best": 0.3285531557816188,
246
+ "expert_success_rate": 0.8584070796460177,
247
+ "num_groups": 113,
248
+ "oracle_success_rate": 0.9380530973451328,
249
+ "policy_expert_regret": 0.9874511470573138,
250
+ "policy_oracle_regret": 1.0666533279999169,
251
+ "policy_rollout_progress": 0.6354583977861742,
252
+ "policy_rollout_success_rate": 0.23008849557522124,
253
+ "restore_max_error": 2.384185791015625e-07
254
+ },
255
+ "PickCube-v1": {
256
+ "action_mse_to_best": 0.3225268117638062,
257
+ "expert_success_rate": 0.9402173913043478,
258
+ "num_groups": 184,
259
+ "oracle_success_rate": 0.9456521739130435,
260
+ "policy_expert_regret": 1.0648617831807907,
261
+ "policy_oracle_regret": 1.064885787655721,
262
+ "policy_rollout_progress": 0.5749052834352884,
263
+ "policy_rollout_success_rate": 0.29891304347826086,
264
+ "restore_max_error": 2.384185791015625e-07
265
+ },
266
+ "PullCube-v1": {
267
+ "action_mse_to_best": 0.5953935721006832,
268
+ "expert_success_rate": 0.25,
269
+ "num_groups": 76,
270
+ "oracle_success_rate": 0.40789473684210525,
271
+ "policy_expert_regret": 0.2746618644726512,
272
+ "policy_oracle_regret": 0.49728695281144036,
273
+ "policy_rollout_progress": 0.3028531574742182,
274
+ "policy_rollout_success_rate": 0.19736842105263158,
275
+ "restore_max_error": 2.3096799850463867e-07
276
+ },
277
+ "PushCube-v1": {
278
+ "action_mse_to_best": 0.3595503710948669,
279
+ "expert_success_rate": 0.8198198198198198,
280
+ "num_groups": 111,
281
+ "oracle_success_rate": 1.0,
282
+ "policy_expert_regret": 0.32904101532321794,
283
+ "policy_oracle_regret": 0.34620858245604746,
284
+ "policy_rollout_progress": 0.8339715977241328,
285
+ "policy_rollout_success_rate": 0.8198198198198198,
286
+ "restore_max_error": 1.7881393432617188e-07
287
+ },
288
+ "StackCube-v1": {
289
+ "action_mse_to_best": 0.526873398326583,
290
+ "expert_success_rate": 0.6923076923076923,
291
+ "num_groups": 91,
292
+ "oracle_success_rate": 0.8571428571428571,
293
+ "policy_expert_regret": 1.1327610128855967,
294
+ "policy_oracle_regret": 1.292421804188372,
295
+ "policy_rollout_progress": 0.3981835229711218,
296
+ "policy_rollout_success_rate": 0.15384615384615385,
297
+ "restore_max_error": 1.9371509552001953e-07
298
+ }
299
+ }
300
+ },
301
+ {
302
+ "seed": 2,
303
+ "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_nmchallenger0p015.json",
304
+ "num_groups": 575,
305
+ "selection_mode": "retrieval_residual",
306
+ "num_candidates": 48,
307
+ "candidate_sigma": 0.0,
308
+ "selection_margin": 0.2,
309
+ "prepend_policy_candidate": false,
310
+ "field_optim_steps": 0,
311
+ "field_optim_step_size": 0.0,
312
+ "field_optim_trust_radius": 0.0,
313
+ "field_optim_l2_penalty": 0.0,
314
+ "retrieval_neighbors": 4,
315
+ "retrieval_metric": "raw",
316
+ "retrieval_type_min_success": 0.0,
317
+ "retrieval_type_success_bonus_scale": 0.0,
318
+ "retrieval_residual_consensus_penalty_scale": 0.0,
319
+ "retrieval_residual_min_source_progress": 0.0,
320
+ "retrieval_residual_min_source_advantage": -1000000000.0,
321
+ "retrieval_residual_source_progress_bonus_scale": 0.0,
322
+ "retrieval_residual_source_score_bonus_scale": 0.0,
323
+ "retrieval_residual_source_advantage_bonus_scale": 0.0,
324
+ "retrieval_residual_composite_l2_penalty_scale": 0.0,
325
+ "retrieval_residual_scale": 1.0,
326
+ "retrieval_residual_scales": [
327
+ 0.35,
328
+ 0.4,
329
+ 0.45
330
+ ],
331
+ "retrieval_residual_anchor": "expert",
332
+ "retrieval_residual_direction": "candidate_minus_anchor",
333
+ "retrieval_residual_reduce": "compose_mean_by_type",
334
+ "retrieval_residual_challenger_types": [
335
+ "residual_near_miss"
336
+ ],
337
+ "retrieval_residual_challenger_margin": 0.015,
338
+ "lattice_exclude_types": [
339
+ "residual_random_negative",
340
+ "residual_wrong_direction",
341
+ "residual_near_miss+residual_no_op"
342
+ ],
343
+ "candidate_type_bonuses": {
344
+ "residual_no_op": 0.03
345
+ },
346
+ "candidate_type_bonus_components": false,
347
+ "candidate_oracle_rollouts": 0,
348
+ "candidate_oracle_unique_tolerance": 1e-06,
349
+ "candidate_oracle_success_rate": null,
350
+ "candidate_oracle_progress": null,
351
+ "candidate_oracle_score": null,
352
+ "candidate_oracle_regret": null,
353
+ "candidate_oracle_score_gain_over_selected": null,
354
+ "candidate_oracle_unique_count": null,
355
+ "candidate_oracle_improvement_rate": null,
356
+ "candidate_oracle_selected_branch_success_rate": null,
357
+ "candidate_oracle_selected_branch_progress": null,
358
+ "candidate_oracle_type_counts": {},
359
+ "candidate_oracle_best_branch_rank": null,
360
+ "candidate_oracle_best_branch_rank_counts": {},
361
+ "candidate_oracle_branch_success_rates": [],
362
+ "candidate_oracle_branch_progress": [],
363
+ "candidate_oracle_branch_score_gains_over_selected": [],
364
+ "selected_residual_scale_counts": {
365
+ "0.35": 521,
366
+ "0.45": 46,
367
+ "0.4": 8
368
+ },
369
+ "policy_rollout_success_rate": 0.3739130434782609,
370
+ "policy_rollout_progress": 0.5890178841644007,
371
+ "oracle_success_rate": 0.8765217391304347,
372
+ "action_mse_to_best": 0.4279105448682347,
373
+ "best_policy_val": {
374
+ "bc_loss": 0.11367896075050037,
375
+ "field_effect_loss": 0.009670218582161598,
376
+ "field_potential_loss": 0.2641640139950646,
377
+ "field_preference_loss": 0.5130490180518892,
378
+ "lattice_edges": 3833.3333333333335,
379
+ "progress_mae": 0.2021729110015763,
380
+ "rank_acc": 0.8333857821093665,
381
+ "rank_loss": 0.5130119257503085,
382
+ "regret_mae": 0.3958987047274907,
383
+ "success_accuracy": 0.8680730561415354,
384
+ "total_loss": 1.4394984311527677
385
+ },
386
+ "per_task": {
387
+ "LiftPegUpright-v1": {
388
+ "action_mse_to_best": 0.35602021238203935,
389
+ "expert_success_rate": 0.8229166666666666,
390
+ "num_groups": 96,
391
+ "oracle_success_rate": 0.9270833333333334,
392
+ "policy_expert_regret": 0.8475987246880928,
393
+ "policy_oracle_regret": 0.9538775502393643,
394
+ "policy_rollout_progress": 0.6463976396868626,
395
+ "policy_rollout_success_rate": 0.3333333333333333,
396
+ "restore_max_error": 2.384185791015625e-07
397
+ },
398
+ "PickCube-v1": {
399
+ "action_mse_to_best": 0.33907994780350814,
400
+ "expert_success_rate": 0.9444444444444444,
401
+ "num_groups": 198,
402
+ "oracle_success_rate": 0.9595959595959596,
403
+ "policy_expert_regret": 0.9807589900952698,
404
+ "policy_oracle_regret": 0.9878424406239782,
405
+ "policy_rollout_progress": 0.6357297236382058,
406
+ "policy_rollout_success_rate": 0.3383838383838384,
407
+ "restore_max_error": 2.384185791015625e-07
408
+ },
409
+ "PullCube-v1": {
410
+ "action_mse_to_best": 0.6399203074475129,
411
+ "expert_success_rate": 0.24444444444444444,
412
+ "num_groups": 90,
413
+ "oracle_success_rate": 0.4666666666666667,
414
+ "policy_expert_regret": 0.269801827851269,
415
+ "policy_oracle_regret": 0.47522657581915456,
416
+ "policy_rollout_progress": 0.36099742491626075,
417
+ "policy_rollout_success_rate": 0.25555555555555554,
418
+ "restore_max_error": 2.384185791015625e-07
419
+ },
420
+ "PushCube-v1": {
421
+ "action_mse_to_best": 0.3970492488824495,
422
+ "expert_success_rate": 0.8514851485148515,
423
+ "num_groups": 101,
424
+ "oracle_success_rate": 1.0,
425
+ "policy_expert_regret": 0.3425388766987489,
426
+ "policy_oracle_regret": 0.39744112810285964,
427
+ "policy_rollout_progress": 0.8104796639763483,
428
+ "policy_rollout_success_rate": 0.7920792079207921,
429
+ "restore_max_error": 2.384185791015625e-07
430
+ },
431
+ "StackCube-v1": {
432
+ "action_mse_to_best": 0.5226443493117888,
433
+ "expert_success_rate": 0.7666666666666667,
434
+ "num_groups": 90,
435
+ "oracle_success_rate": 0.9111111111111111,
436
+ "policy_expert_regret": 1.1750739713509877,
437
+ "policy_oracle_regret": 1.3371944083107843,
438
+ "policy_rollout_progress": 0.4045378933350245,
439
+ "policy_rollout_success_rate": 0.14444444444444443,
440
+ "restore_max_error": 2.086162567138672e-07
441
+ }
442
+ }
443
+ }
444
+ ]
445
+ }
results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmchallenger0p015_summary.md ADDED
@@ -0,0 +1,19 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # h=16 Best-Policy Checkpoint Rollout
2
+
3
+ Run root: `/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs`
4
+ Objective: `near_miss_policy_bc5`
5
+ Result file: `policy_rollout_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmchallenger0p015.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: 36.00% +/- 1.25%
11
+ Gain vs h=16 rank checkpoint: +6.26%
12
+ Mean progress: 57.38%
13
+ Mean action MSE to best: 0.407
14
+
15
+ | seed | mode | k | policy cand | retrieval K | retrieval metric | residual anchor | residual direction | residual reduce | min type success | type success bonus | consensus penalty | min source progress | source progress bonus | source score bonus | residual scale | residual scales | margin | sigma | opt steps | trust | success | progress | oracle | candidate oracle | oracle gain | action MSE |
16
+ |---:|---|---:|---|---:|---|---|---|---|---:|---:|---:|---:|---:|---:|---:|---|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|
17
+ | 0 | retrieval_residual | 48 | no | 4 | raw | expert | candidate_minus_anchor | compose_mean_by_type + challenger residual_near_miss@0.01 | 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 | 35.65% | 55.96% | 85.74% | n/a | n/a | 0.395 |
18
+ | 1 | retrieval_residual | 48 | no | 4 | raw | expert | candidate_minus_anchor | compose_mean_by_type + challenger residual_near_miss@0.01 | 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 | 34.96% | 57.29% | 86.96% | n/a | n/a | 0.399 |
19
+ | 2 | retrieval_residual | 48 | no | 4 | raw | expert | candidate_minus_anchor | compose_mean_by_type + challenger residual_near_miss@0.01 | 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 | 37.39% | 58.90% | 87.65% | n/a | n/a | 0.428 |