anhtld commited on
Commit
8ae17f4
·
verified ·
1 Parent(s): 3730407

Auto-sync: 2026-06-29 02:57:43 (part 2)

Browse files
results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_grid040045050_safe_margin0p20_mean_by_type_noopbonus0p03_summary.json ADDED
@@ -0,0 +1,346 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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_grid040045050_safe_margin0p20_mean_by_type_noopbonus0p03.json",
5
+ "num_completed": 3,
6
+ "baseline_h4_policy_success": 0.2967,
7
+ "baseline_h16_rank_checkpoint_success": 0.29739130434782607,
8
+ "mean_success": 0.3536231884057971,
9
+ "std_success": 0.010190374394925756,
10
+ "mean_progress": 0.5691949463564601,
11
+ "mean_action_mse_to_best": 0.3983363078623686,
12
+ "gain_vs_h4": 0.056923188405797076,
13
+ "gain_vs_h16_rank_checkpoint": 0.05623188405797103,
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_grid040045050_safe_margin0p20_mean_by_type_noopbonus0p03.json",
18
+ "num_groups": 575,
19
+ "selection_mode": "retrieval_residual",
20
+ "num_candidates": 18,
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": 0.0,
34
+ "retrieval_residual_scale": 0.4,
35
+ "retrieval_residual_scales": [
36
+ 0.4,
37
+ 0.45,
38
+ 0.5
39
+ ],
40
+ "retrieval_residual_anchor": "expert",
41
+ "retrieval_residual_reduce": "mean_by_type",
42
+ "candidate_type_bonuses": {
43
+ "residual_no_op": 0.03
44
+ },
45
+ "selected_residual_scale_counts": {
46
+ "0.4": 543,
47
+ "0.5": 30,
48
+ "0.45": 2
49
+ },
50
+ "policy_rollout_success_rate": 0.34608695652173915,
51
+ "policy_rollout_progress": 0.5553471852506956,
52
+ "oracle_success_rate": 0.8573913043478261,
53
+ "action_mse_to_best": 0.3837931162734394,
54
+ "best_policy_val": {
55
+ "bc_loss": 0.13721593966086706,
56
+ "field_effect_loss": 0.009290305380192068,
57
+ "field_potential_loss": 0.2666468388504452,
58
+ "field_preference_loss": 0.5130573478009965,
59
+ "lattice_edges": 3833.3333333333335,
60
+ "progress_mae": 0.1933159919248687,
61
+ "rank_acc": 0.8265031774838766,
62
+ "rank_loss": 0.5130523675017886,
63
+ "regret_mae": 0.3756548762321472,
64
+ "success_accuracy": 0.8773836526605818,
65
+ "total_loss": 1.5581054819954767
66
+ },
67
+ "per_task": {
68
+ "LiftPegUpright-v1": {
69
+ "action_mse_to_best": 0.31969772460726426,
70
+ "expert_success_rate": 0.8865979381443299,
71
+ "num_groups": 97,
72
+ "oracle_success_rate": 0.9278350515463918,
73
+ "policy_expert_regret": 1.0783334703175063,
74
+ "policy_oracle_regret": 1.1047489480259491,
75
+ "policy_rollout_progress": 0.5714634232914325,
76
+ "policy_rollout_success_rate": 0.2268041237113402,
77
+ "restore_max_error": 2.384185791015625e-07
78
+ },
79
+ "PickCube-v1": {
80
+ "action_mse_to_best": 0.3028640510299458,
81
+ "expert_success_rate": 0.9375,
82
+ "num_groups": 208,
83
+ "oracle_success_rate": 0.9471153846153846,
84
+ "policy_expert_regret": 0.9681451314916978,
85
+ "policy_oracle_regret": 0.9815679989182032,
86
+ "policy_rollout_progress": 0.6139292046427727,
87
+ "policy_rollout_success_rate": 0.33653846153846156,
88
+ "restore_max_error": 2.384185791015625e-07
89
+ },
90
+ "PullCube-v1": {
91
+ "action_mse_to_best": 0.6037803367747889,
92
+ "expert_success_rate": 0.19480519480519481,
93
+ "num_groups": 77,
94
+ "oracle_success_rate": 0.36363636363636365,
95
+ "policy_expert_regret": 0.24448560403853462,
96
+ "policy_oracle_regret": 0.41804164213237244,
97
+ "policy_rollout_progress": 0.31177073029510965,
98
+ "policy_rollout_success_rate": 0.2077922077922078,
99
+ "restore_max_error": 2.3096799850463867e-07
100
+ },
101
+ "PushCube-v1": {
102
+ "action_mse_to_best": 0.36579142193844977,
103
+ "expert_success_rate": 0.8617021276595744,
104
+ "num_groups": 94,
105
+ "oracle_success_rate": 0.9893617021276596,
106
+ "policy_expert_regret": 0.5508279580068081,
107
+ "policy_oracle_regret": 0.6283786044792926,
108
+ "policy_rollout_progress": 0.7014086295632606,
109
+ "policy_rollout_success_rate": 0.6702127659574468,
110
+ "restore_max_error": 2.384185791015625e-07
111
+ },
112
+ "StackCube-v1": {
113
+ "action_mse_to_best": 0.4626177814365788,
114
+ "expert_success_rate": 0.6767676767676768,
115
+ "num_groups": 99,
116
+ "oracle_success_rate": 0.8585858585858586,
117
+ "policy_expert_regret": 0.9198090114978829,
118
+ "policy_oracle_regret": 1.0823988696240416,
119
+ "policy_rollout_progress": 0.46723886345974125,
120
+ "policy_rollout_success_rate": 0.2828282828282828,
121
+ "restore_max_error": 2.384185791015625e-07
122
+ }
123
+ }
124
+ },
125
+ {
126
+ "seed": 1,
127
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/near_miss_policy_bc5/seed_1/policy_rollout_retrieval_residual_k4_grid040045050_safe_margin0p20_mean_by_type_noopbonus0p03.json",
128
+ "num_groups": 575,
129
+ "selection_mode": "retrieval_residual",
130
+ "num_candidates": 18,
131
+ "candidate_sigma": 0.0,
132
+ "selection_margin": 0.2,
133
+ "prepend_policy_candidate": false,
134
+ "field_optim_steps": 0,
135
+ "field_optim_step_size": 0.0,
136
+ "field_optim_trust_radius": 0.0,
137
+ "field_optim_l2_penalty": 0.0,
138
+ "retrieval_neighbors": 4,
139
+ "retrieval_metric": "raw",
140
+ "retrieval_type_min_success": 0.0,
141
+ "retrieval_residual_min_source_progress": 0.0,
142
+ "retrieval_residual_source_progress_bonus_scale": 0.0,
143
+ "retrieval_residual_source_score_bonus_scale": 0.0,
144
+ "retrieval_residual_scale": 0.4,
145
+ "retrieval_residual_scales": [
146
+ 0.4,
147
+ 0.45,
148
+ 0.5
149
+ ],
150
+ "retrieval_residual_anchor": "expert",
151
+ "retrieval_residual_reduce": "mean_by_type",
152
+ "candidate_type_bonuses": {
153
+ "residual_no_op": 0.03
154
+ },
155
+ "selected_residual_scale_counts": {
156
+ "0.4": 549,
157
+ "0.45": 4,
158
+ "0.5": 22
159
+ },
160
+ "policy_rollout_success_rate": 0.34956521739130436,
161
+ "policy_rollout_progress": 0.5711554712316265,
162
+ "oracle_success_rate": 0.8695652173913043,
163
+ "action_mse_to_best": 0.39108053201080667,
164
+ "best_policy_val": {
165
+ "bc_loss": 0.1315989200439718,
166
+ "field_effect_loss": 0.010110370814800262,
167
+ "field_potential_loss": 0.31546536915832096,
168
+ "field_preference_loss": 0.4960637440284093,
169
+ "lattice_edges": 3833.3333333333335,
170
+ "progress_mae": 0.19023226698239645,
171
+ "rank_acc": 0.8255469501018524,
172
+ "rank_loss": 0.4960531195004781,
173
+ "regret_mae": 0.42156532737943864,
174
+ "success_accuracy": 0.8807473679383596,
175
+ "total_loss": 1.561984618504842
176
+ },
177
+ "per_task": {
178
+ "LiftPegUpright-v1": {
179
+ "action_mse_to_best": 0.3255710795406351,
180
+ "expert_success_rate": 0.8584070796460177,
181
+ "num_groups": 113,
182
+ "oracle_success_rate": 0.9380530973451328,
183
+ "policy_expert_regret": 0.9873926745051831,
184
+ "policy_oracle_regret": 1.0665947045900126,
185
+ "policy_rollout_progress": 0.6355170211960784,
186
+ "policy_rollout_success_rate": 0.23008849557522124,
187
+ "restore_max_error": 2.384185791015625e-07
188
+ },
189
+ "PickCube-v1": {
190
+ "action_mse_to_best": 0.32045821142990305,
191
+ "expert_success_rate": 0.9402173913043478,
192
+ "num_groups": 184,
193
+ "oracle_success_rate": 0.9456521739130435,
194
+ "policy_expert_regret": 1.055350362100517,
195
+ "policy_oracle_regret": 1.0553743665754471,
196
+ "policy_rollout_progress": 0.5789819219068665,
197
+ "policy_rollout_success_rate": 0.30434782608695654,
198
+ "restore_max_error": 2.384185791015625e-07
199
+ },
200
+ "PullCube-v1": {
201
+ "action_mse_to_best": 0.573825633937591,
202
+ "expert_success_rate": 0.25,
203
+ "num_groups": 76,
204
+ "oracle_success_rate": 0.40789473684210525,
205
+ "policy_expert_regret": 0.3001050222174902,
206
+ "policy_oracle_regret": 0.522557810194006,
207
+ "policy_rollout_progress": 0.29004565131311355,
208
+ "policy_rollout_success_rate": 0.18421052631578946,
209
+ "restore_max_error": 2.384185791015625e-07
210
+ },
211
+ "PushCube-v1": {
212
+ "action_mse_to_best": 0.35999264198917524,
213
+ "expert_success_rate": 0.8198198198198198,
214
+ "num_groups": 111,
215
+ "oracle_success_rate": 1.0,
216
+ "policy_expert_regret": 0.3294863182682175,
217
+ "policy_oracle_regret": 0.346653885401047,
218
+ "policy_rollout_progress": 0.8335262947791332,
219
+ "policy_rollout_success_rate": 0.8198198198198198,
220
+ "restore_max_error": 2.3795291781425476e-07
221
+ },
222
+ "StackCube-v1": {
223
+ "action_mse_to_best": 0.5005223249996101,
224
+ "expert_success_rate": 0.6923076923076923,
225
+ "num_groups": 91,
226
+ "oracle_success_rate": 0.8571428571428571,
227
+ "policy_expert_regret": 1.140796109542742,
228
+ "policy_oracle_regret": 1.3004579566992247,
229
+ "policy_rollout_progress": 0.3901473704602692,
230
+ "policy_rollout_success_rate": 0.15384615384615385,
231
+ "restore_max_error": 1.9371509552001953e-07
232
+ }
233
+ }
234
+ },
235
+ {
236
+ "seed": 2,
237
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/near_miss_policy_bc5/seed_2/policy_rollout_retrieval_residual_k4_grid040045050_safe_margin0p20_mean_by_type_noopbonus0p03.json",
238
+ "num_groups": 575,
239
+ "selection_mode": "retrieval_residual",
240
+ "num_candidates": 18,
241
+ "candidate_sigma": 0.0,
242
+ "selection_margin": 0.2,
243
+ "prepend_policy_candidate": false,
244
+ "field_optim_steps": 0,
245
+ "field_optim_step_size": 0.0,
246
+ "field_optim_trust_radius": 0.0,
247
+ "field_optim_l2_penalty": 0.0,
248
+ "retrieval_neighbors": 4,
249
+ "retrieval_metric": "raw",
250
+ "retrieval_type_min_success": 0.0,
251
+ "retrieval_residual_min_source_progress": 0.0,
252
+ "retrieval_residual_source_progress_bonus_scale": 0.0,
253
+ "retrieval_residual_source_score_bonus_scale": 0.0,
254
+ "retrieval_residual_scale": 0.4,
255
+ "retrieval_residual_scales": [
256
+ 0.4,
257
+ 0.45,
258
+ 0.5
259
+ ],
260
+ "retrieval_residual_anchor": "expert",
261
+ "retrieval_residual_reduce": "mean_by_type",
262
+ "candidate_type_bonuses": {
263
+ "residual_no_op": 0.03
264
+ },
265
+ "selected_residual_scale_counts": {
266
+ "0.4": 549,
267
+ "0.45": 6,
268
+ "0.5": 20
269
+ },
270
+ "policy_rollout_success_rate": 0.3652173913043478,
271
+ "policy_rollout_progress": 0.581082182587058,
272
+ "oracle_success_rate": 0.8765217391304347,
273
+ "action_mse_to_best": 0.42013527530285977,
274
+ "best_policy_val": {
275
+ "bc_loss": 0.11367896075050037,
276
+ "field_effect_loss": 0.009670218582161598,
277
+ "field_potential_loss": 0.2641640139950646,
278
+ "field_preference_loss": 0.5130490180518892,
279
+ "lattice_edges": 3833.3333333333335,
280
+ "progress_mae": 0.2021729110015763,
281
+ "rank_acc": 0.8333857821093665,
282
+ "rank_loss": 0.5130119257503085,
283
+ "regret_mae": 0.3958987047274907,
284
+ "success_accuracy": 0.8680730561415354,
285
+ "total_loss": 1.4394984311527677
286
+ },
287
+ "per_task": {
288
+ "LiftPegUpright-v1": {
289
+ "action_mse_to_best": 0.35442226944239036,
290
+ "expert_success_rate": 0.8229166666666666,
291
+ "num_groups": 96,
292
+ "oracle_success_rate": 0.9270833333333334,
293
+ "policy_expert_regret": 0.8476849501021206,
294
+ "policy_oracle_regret": 0.9539546399998168,
295
+ "policy_rollout_progress": 0.6463205499264101,
296
+ "policy_rollout_success_rate": 0.3333333333333333,
297
+ "restore_max_error": 1.955777406692505e-07
298
+ },
299
+ "PickCube-v1": {
300
+ "action_mse_to_best": 0.3370037544518709,
301
+ "expert_success_rate": 0.9444444444444444,
302
+ "num_groups": 198,
303
+ "oracle_success_rate": 0.9595959595959596,
304
+ "policy_expert_regret": 0.9807443945113608,
305
+ "policy_oracle_regret": 0.987827845040069,
306
+ "policy_rollout_progress": 0.6357443192221149,
307
+ "policy_rollout_success_rate": 0.3383838383838384,
308
+ "restore_max_error": 2.384185791015625e-07
309
+ },
310
+ "PullCube-v1": {
311
+ "action_mse_to_best": 0.6273176141911083,
312
+ "expert_success_rate": 0.24444444444444444,
313
+ "num_groups": 90,
314
+ "oracle_success_rate": 0.4666666666666667,
315
+ "policy_expert_regret": 0.2938225742705375,
316
+ "policy_oracle_regret": 0.5421275411284013,
317
+ "policy_rollout_progress": 0.31943315598010585,
318
+ "policy_rollout_success_rate": 0.2111111111111111,
319
+ "restore_max_error": 2.384185791015625e-07
320
+ },
321
+ "PushCube-v1": {
322
+ "action_mse_to_best": 0.39170166304205906,
323
+ "expert_success_rate": 0.8514851485148515,
324
+ "num_groups": 101,
325
+ "oracle_success_rate": 1.0,
326
+ "policy_expert_regret": 0.3976949640724919,
327
+ "policy_oracle_regret": 0.43470467125425244,
328
+ "policy_rollout_progress": 0.7930181010229753,
329
+ "policy_rollout_success_rate": 0.7722772277227723,
330
+ "restore_max_error": 2.384185791015625e-07
331
+ },
332
+ "StackCube-v1": {
333
+ "action_mse_to_best": 0.4978449867417415,
334
+ "expert_success_rate": 0.7666666666666667,
335
+ "num_groups": 90,
336
+ "oracle_success_rate": 0.9111111111111111,
337
+ "policy_expert_regret": 1.162587201760875,
338
+ "policy_oracle_regret": 1.311838612291548,
339
+ "policy_rollout_progress": 0.4150477194123798,
340
+ "policy_rollout_success_rate": 0.15555555555555556,
341
+ "restore_max_error": 2.2351741790771484e-07
342
+ }
343
+ }
344
+ }
345
+ ]
346
+ }
results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_grid040045050_safe_margin0p20_mean_by_type_noopbonus0p03_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_grid040045050_safe_margin0p20_mean_by_type_noopbonus0p03.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: 35.36% +/- 1.02%
11
+ Gain vs h=16 rank checkpoint: +5.62%
12
+ Mean progress: 56.92%
13
+ Mean action MSE to best: 0.398
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 | 18 | no | 4 | raw | expert | mean_by_type | 0.00 | 0.00 | 0.000 | 0.000 | 0.40 | 0.40,0.45,0.50 | 0.200 | 0.00 | 0 | 0.00 | 34.61% | 55.53% | 85.74% | 0.384 |
18
+ | 1 | retrieval_residual | 18 | no | 4 | raw | expert | mean_by_type | 0.00 | 0.00 | 0.000 | 0.000 | 0.40 | 0.40,0.45,0.50 | 0.200 | 0.00 | 0 | 0.00 | 34.96% | 57.12% | 86.96% | 0.391 |
19
+ | 2 | retrieval_residual | 18 | no | 4 | raw | expert | mean_by_type | 0.00 | 0.00 | 0.000 | 0.000 | 0.40 | 0.40,0.45,0.50 | 0.200 | 0.00 | 0 | 0.00 | 36.52% | 58.11% | 87.65% | 0.420 |