anhtld commited on
Commit
81fee44
·
verified ·
1 Parent(s): 72bd06c

Auto-sync: 2026-06-29 01:47:15

Browse files
results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4s040_safe_margin0p20_mean_by_type_srcscorebonus0p015_summary.json ADDED
@@ -0,0 +1,322 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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_k4s040_safe_margin0p20_mean_by_type_srcscorebonus0p015.json",
5
+ "num_completed": 3,
6
+ "baseline_h4_policy_success": 0.2967,
7
+ "baseline_h16_rank_checkpoint_success": 0.29739130434782607,
8
+ "mean_success": 0.35246376811594204,
9
+ "std_success": 0.014164396200429698,
10
+ "mean_progress": 0.5668091635216157,
11
+ "mean_action_mse_to_best": 0.3954103358281587,
12
+ "gain_vs_h4": 0.055763768115942025,
13
+ "gain_vs_h16_rank_checkpoint": 0.055072463768115976,
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_k4s040_safe_margin0p20_mean_by_type_srcscorebonus0p015.json",
18
+ "num_groups": 575,
19
+ "selection_mode": "retrieval_residual",
20
+ "num_candidates": 6,
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_residual_min_source_progress": 0.0,
32
+ "retrieval_residual_source_progress_bonus_scale": 0.0,
33
+ "retrieval_residual_source_score_bonus_scale": 0.015,
34
+ "retrieval_residual_scale": 0.4,
35
+ "retrieval_residual_scales": [],
36
+ "retrieval_residual_anchor": "expert",
37
+ "retrieval_residual_reduce": "mean_by_type",
38
+ "candidate_type_bonuses": {},
39
+ "selected_residual_scale_counts": {
40
+ "0.4": 575
41
+ },
42
+ "policy_rollout_success_rate": 0.34608695652173915,
43
+ "policy_rollout_progress": 0.5509786044636175,
44
+ "oracle_success_rate": 0.8573913043478261,
45
+ "action_mse_to_best": 0.3816270524878865,
46
+ "best_policy_val": {
47
+ "bc_loss": 0.13721593966086706,
48
+ "field_effect_loss": 0.009290305380192068,
49
+ "field_potential_loss": 0.2666468388504452,
50
+ "field_preference_loss": 0.5130573478009965,
51
+ "lattice_edges": 3833.3333333333335,
52
+ "progress_mae": 0.1933159919248687,
53
+ "rank_acc": 0.8265031774838766,
54
+ "rank_loss": 0.5130523675017886,
55
+ "regret_mae": 0.3756548762321472,
56
+ "success_accuracy": 0.8773836526605818,
57
+ "total_loss": 1.5581054819954767
58
+ },
59
+ "per_task": {
60
+ "LiftPegUpright-v1": {
61
+ "action_mse_to_best": 0.31969772460726426,
62
+ "expert_success_rate": 0.8865979381443299,
63
+ "num_groups": 97,
64
+ "oracle_success_rate": 0.9278350515463918,
65
+ "policy_expert_regret": 1.0783334703175063,
66
+ "policy_oracle_regret": 1.1047489480259491,
67
+ "policy_rollout_progress": 0.5714634232914325,
68
+ "policy_rollout_success_rate": 0.2268041237113402,
69
+ "restore_max_error": 2.384185791015625e-07
70
+ },
71
+ "PickCube-v1": {
72
+ "action_mse_to_best": 0.30265718372538686,
73
+ "expert_success_rate": 0.9375,
74
+ "num_groups": 208,
75
+ "oracle_success_rate": 0.9471153846153846,
76
+ "policy_expert_regret": 0.9612012499800096,
77
+ "policy_oracle_regret": 0.974624117406515,
78
+ "policy_rollout_progress": 0.6160653938467686,
79
+ "policy_rollout_success_rate": 0.34134615384615385,
80
+ "restore_max_error": 2.384185791015625e-07
81
+ },
82
+ "PullCube-v1": {
83
+ "action_mse_to_best": 0.6002476956550177,
84
+ "expert_success_rate": 0.19480519480519481,
85
+ "num_groups": 77,
86
+ "oracle_success_rate": 0.36363636363636365,
87
+ "policy_expert_regret": 0.246578575468731,
88
+ "policy_oracle_regret": 0.4201346135625688,
89
+ "policy_rollout_progress": 0.3096777588649133,
90
+ "policy_rollout_success_rate": 0.2077922077922078,
91
+ "restore_max_error": 2.3096799850463867e-07
92
+ },
93
+ "PushCube-v1": {
94
+ "action_mse_to_best": 0.36465501579198434,
95
+ "expert_success_rate": 0.8617021276595744,
96
+ "num_groups": 94,
97
+ "oracle_success_rate": 0.9893617021276596,
98
+ "policy_expert_regret": 0.5508309629686335,
99
+ "policy_oracle_regret": 0.647163599729538,
100
+ "policy_rollout_progress": 0.6932619321853557,
101
+ "policy_rollout_success_rate": 0.6595744680851063,
102
+ "restore_max_error": 2.384185791015625e-07
103
+ },
104
+ "StackCube-v1": {
105
+ "action_mse_to_best": 0.45429835988754275,
106
+ "expert_success_rate": 0.6767676767676768,
107
+ "num_groups": 99,
108
+ "oracle_success_rate": 0.8585858585858586,
109
+ "policy_expert_regret": 0.9403071221378114,
110
+ "policy_oracle_regret": 1.10289698026397,
111
+ "policy_rollout_progress": 0.4467407528198127,
112
+ "policy_rollout_success_rate": 0.2828282828282828,
113
+ "restore_max_error": 2.384185791015625e-07
114
+ }
115
+ }
116
+ },
117
+ {
118
+ "seed": 1,
119
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/near_miss_policy_bc5/seed_1/policy_rollout_retrieval_residual_k4s040_safe_margin0p20_mean_by_type_srcscorebonus0p015.json",
120
+ "num_groups": 575,
121
+ "selection_mode": "retrieval_residual",
122
+ "num_candidates": 6,
123
+ "candidate_sigma": 0.0,
124
+ "selection_margin": 0.2,
125
+ "prepend_policy_candidate": false,
126
+ "field_optim_steps": 0,
127
+ "field_optim_step_size": 0.0,
128
+ "field_optim_trust_radius": 0.0,
129
+ "field_optim_l2_penalty": 0.0,
130
+ "retrieval_neighbors": 4,
131
+ "retrieval_metric": "raw",
132
+ "retrieval_type_min_success": 0.0,
133
+ "retrieval_residual_min_source_progress": 0.0,
134
+ "retrieval_residual_source_progress_bonus_scale": 0.0,
135
+ "retrieval_residual_source_score_bonus_scale": 0.015,
136
+ "retrieval_residual_scale": 0.4,
137
+ "retrieval_residual_scales": [],
138
+ "retrieval_residual_anchor": "expert",
139
+ "retrieval_residual_reduce": "mean_by_type",
140
+ "candidate_type_bonuses": {},
141
+ "selected_residual_scale_counts": {
142
+ "0.4": 575
143
+ },
144
+ "policy_rollout_success_rate": 0.3426086956521739,
145
+ "policy_rollout_progress": 0.5665316749817652,
146
+ "oracle_success_rate": 0.8695652173913043,
147
+ "action_mse_to_best": 0.38801324410445015,
148
+ "best_policy_val": {
149
+ "bc_loss": 0.1315989200439718,
150
+ "field_effect_loss": 0.010110370814800262,
151
+ "field_potential_loss": 0.31546536915832096,
152
+ "field_preference_loss": 0.4960637440284093,
153
+ "lattice_edges": 3833.3333333333335,
154
+ "progress_mae": 0.19023226698239645,
155
+ "rank_acc": 0.8255469501018524,
156
+ "rank_loss": 0.4960531195004781,
157
+ "regret_mae": 0.42156532737943864,
158
+ "success_accuracy": 0.8807473679383596,
159
+ "total_loss": 1.561984618504842
160
+ },
161
+ "per_task": {
162
+ "LiftPegUpright-v1": {
163
+ "action_mse_to_best": 0.3255710795406351,
164
+ "expert_success_rate": 0.8584070796460177,
165
+ "num_groups": 113,
166
+ "oracle_success_rate": 0.9380530973451328,
167
+ "policy_expert_regret": 0.9873926745051831,
168
+ "policy_oracle_regret": 1.0665947045900126,
169
+ "policy_rollout_progress": 0.6355170211960784,
170
+ "policy_rollout_success_rate": 0.23008849557522124,
171
+ "restore_max_error": 2.384185791015625e-07
172
+ },
173
+ "PickCube-v1": {
174
+ "action_mse_to_best": 0.31969962222501636,
175
+ "expert_success_rate": 0.9402173913043478,
176
+ "num_groups": 184,
177
+ "oracle_success_rate": 0.9456521739130435,
178
+ "policy_expert_regret": 1.0709032043308506,
179
+ "policy_oracle_regret": 1.0709272088057806,
180
+ "policy_rollout_progress": 0.5742986448939241,
181
+ "policy_rollout_success_rate": 0.29347826086956524,
182
+ "restore_max_error": 2.384185791015625e-07
183
+ },
184
+ "PullCube-v1": {
185
+ "action_mse_to_best": 0.5715771112591028,
186
+ "expert_success_rate": 0.25,
187
+ "num_groups": 76,
188
+ "oracle_success_rate": 0.40789473684210525,
189
+ "policy_expert_regret": 0.30013844498286124,
190
+ "policy_oracle_regret": 0.5234715515061429,
191
+ "policy_rollout_progress": 0.28913191000097677,
192
+ "policy_rollout_success_rate": 0.18421052631578946,
193
+ "restore_max_error": 2.384185791015625e-07
194
+ },
195
+ "PushCube-v1": {
196
+ "action_mse_to_best": 0.35765871029716356,
197
+ "expert_success_rate": 0.8198198198198198,
198
+ "num_groups": 111,
199
+ "oracle_success_rate": 1.0,
200
+ "policy_expert_regret": 0.32971009934270706,
201
+ "policy_oracle_regret": 0.38007361851296984,
202
+ "policy_rollout_progress": 0.8181245796852283,
203
+ "policy_rollout_success_rate": 0.8018018018018018,
204
+ "restore_max_error": 2.3795291781425476e-07
205
+ },
206
+ "StackCube-v1": {
207
+ "action_mse_to_best": 0.48739973169106704,
208
+ "expert_success_rate": 0.6923076923076923,
209
+ "num_groups": 91,
210
+ "oracle_success_rate": 0.8571428571428571,
211
+ "policy_expert_regret": 1.1409930893025555,
212
+ "policy_oracle_regret": 1.3006549364590383,
213
+ "policy_rollout_progress": 0.38995039070045556,
214
+ "policy_rollout_success_rate": 0.15384615384615385,
215
+ "restore_max_error": 1.9371509552001953e-07
216
+ }
217
+ }
218
+ },
219
+ {
220
+ "seed": 2,
221
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/near_miss_policy_bc5/seed_2/policy_rollout_retrieval_residual_k4s040_safe_margin0p20_mean_by_type_srcscorebonus0p015.json",
222
+ "num_groups": 575,
223
+ "selection_mode": "retrieval_residual",
224
+ "num_candidates": 6,
225
+ "candidate_sigma": 0.0,
226
+ "selection_margin": 0.2,
227
+ "prepend_policy_candidate": false,
228
+ "field_optim_steps": 0,
229
+ "field_optim_step_size": 0.0,
230
+ "field_optim_trust_radius": 0.0,
231
+ "field_optim_l2_penalty": 0.0,
232
+ "retrieval_neighbors": 4,
233
+ "retrieval_metric": "raw",
234
+ "retrieval_type_min_success": 0.0,
235
+ "retrieval_residual_min_source_progress": 0.0,
236
+ "retrieval_residual_source_progress_bonus_scale": 0.0,
237
+ "retrieval_residual_source_score_bonus_scale": 0.015,
238
+ "retrieval_residual_scale": 0.4,
239
+ "retrieval_residual_scales": [],
240
+ "retrieval_residual_anchor": "expert",
241
+ "retrieval_residual_reduce": "mean_by_type",
242
+ "candidate_type_bonuses": {},
243
+ "selected_residual_scale_counts": {
244
+ "0.4": 575
245
+ },
246
+ "policy_rollout_success_rate": 0.36869565217391304,
247
+ "policy_rollout_progress": 0.5829172111194645,
248
+ "oracle_success_rate": 0.8765217391304347,
249
+ "action_mse_to_best": 0.41659071089213956,
250
+ "best_policy_val": {
251
+ "bc_loss": 0.11367896075050037,
252
+ "field_effect_loss": 0.009670218582161598,
253
+ "field_potential_loss": 0.2641640139950646,
254
+ "field_preference_loss": 0.5130490180518892,
255
+ "lattice_edges": 3833.3333333333335,
256
+ "progress_mae": 0.2021729110015763,
257
+ "rank_acc": 0.8333857821093665,
258
+ "rank_loss": 0.5130119257503085,
259
+ "regret_mae": 0.3958987047274907,
260
+ "success_accuracy": 0.8680730561415354,
261
+ "total_loss": 1.4394984311527677
262
+ },
263
+ "per_task": {
264
+ "LiftPegUpright-v1": {
265
+ "action_mse_to_best": 0.35488338287783944,
266
+ "expert_success_rate": 0.8229166666666666,
267
+ "num_groups": 96,
268
+ "oracle_success_rate": 0.9270833333333334,
269
+ "policy_expert_regret": 0.8476849501021206,
270
+ "policy_oracle_regret": 0.9539546399998168,
271
+ "policy_rollout_progress": 0.6463205499264101,
272
+ "policy_rollout_success_rate": 0.3333333333333333,
273
+ "restore_max_error": 1.955777406692505e-07
274
+ },
275
+ "PickCube-v1": {
276
+ "action_mse_to_best": 0.33545475296977195,
277
+ "expert_success_rate": 0.9444444444444444,
278
+ "num_groups": 198,
279
+ "oracle_success_rate": 0.9595959595959596,
280
+ "policy_expert_regret": 0.9806582634572429,
281
+ "policy_oracle_regret": 0.9877417139859512,
282
+ "policy_rollout_progress": 0.6358304502762328,
283
+ "policy_rollout_success_rate": 0.3383838383838384,
284
+ "restore_max_error": 2.384185791015625e-07
285
+ },
286
+ "PullCube-v1": {
287
+ "action_mse_to_best": 0.6283310380246904,
288
+ "expert_success_rate": 0.24444444444444444,
289
+ "num_groups": 90,
290
+ "oracle_success_rate": 0.4666666666666667,
291
+ "policy_expert_regret": 0.3140720418619821,
292
+ "policy_oracle_regret": 0.5426390113998827,
293
+ "policy_rollout_progress": 0.31892168570862445,
294
+ "policy_rollout_success_rate": 0.2111111111111111,
295
+ "restore_max_error": 2.384185791015625e-07
296
+ },
297
+ "PushCube-v1": {
298
+ "action_mse_to_best": 0.3851195577495169,
299
+ "expert_success_rate": 0.8514851485148515,
300
+ "num_groups": 101,
301
+ "oracle_success_rate": 1.0,
302
+ "policy_expert_regret": 0.3620445999768701,
303
+ "policy_oracle_regret": 0.41694685138098087,
304
+ "policy_rollout_progress": 0.8008749307972369,
305
+ "policy_rollout_success_rate": 0.7821782178217822,
306
+ "restore_max_error": 2.384185791015625e-07
307
+ },
308
+ "StackCube-v1": {
309
+ "action_mse_to_best": 0.4844882684863276,
310
+ "expert_success_rate": 0.7666666666666667,
311
+ "num_groups": 90,
312
+ "oracle_success_rate": 0.9111111111111111,
313
+ "policy_expert_regret": 1.1607355892658233,
314
+ "policy_oracle_regret": 1.2970718792743152,
315
+ "policy_rollout_progress": 0.4182763857973946,
316
+ "policy_rollout_success_rate": 0.16666666666666666,
317
+ "restore_max_error": 2.2351741790771484e-07
318
+ }
319
+ }
320
+ }
321
+ ]
322
+ }
results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4s040_safe_margin0p20_mean_by_type_srcscorebonus0p015_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_k4s040_safe_margin0p20_mean_by_type_srcscorebonus0p015.json`
6
+ Completed seeds: 3
7
+ Baseline h=4 policy success: 29.67%
8
+ Baseline h=16 rank-checkpoint success: 29.74%
9
+
10
+ Mean success: 35.25% +/- 1.42%
11
+ Gain vs h=16 rank checkpoint: +5.51%
12
+ Mean progress: 56.68%
13
+ Mean action MSE to best: 0.395
14
+
15
+ | seed | mode | k | policy cand | retrieval K | retrieval metric | residual anchor | residual reduce | min type success | min source progress | source progress bonus | source score bonus | residual scale | residual scales | margin | sigma | opt steps | trust | success | progress | oracle | action MSE |
16
+ |---:|---|---:|---|---:|---|---|---|---:|---:|---:|---:|---:|---|---:|---:|---:|---:|---:|---:|---:|---:|
17
+ | 0 | retrieval_residual | 6 | no | 4 | raw | expert | mean_by_type | 0.00 | 0.00 | 0.000 | 0.015 | 0.40 | none | 0.200 | 0.00 | 0 | 0.00 | 34.61% | 55.10% | 85.74% | 0.382 |
18
+ | 1 | retrieval_residual | 6 | no | 4 | raw | expert | mean_by_type | 0.00 | 0.00 | 0.000 | 0.015 | 0.40 | none | 0.200 | 0.00 | 0 | 0.00 | 34.26% | 56.65% | 86.96% | 0.388 |
19
+ | 2 | retrieval_residual | 6 | no | 4 | raw | expert | mean_by_type | 0.00 | 0.00 | 0.000 | 0.015 | 0.40 | none | 0.200 | 0.00 | 0 | 0.00 | 36.87% | 58.29% | 87.65% | 0.417 |