anhtld commited on
Commit
218ed13
·
verified ·
1 Parent(s): d8c8d9e

Auto-sync: 2026-06-28 08:21:09

Browse files
results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_scale0p25_safe_types_margin0p20_summary.json ADDED
@@ -0,0 +1,292 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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_scale0p25_safe_types_margin0p20.json",
5
+ "num_completed": 3,
6
+ "baseline_h4_policy_success": 0.2967,
7
+ "baseline_h16_rank_checkpoint_success": 0.29739130434782607,
8
+ "mean_success": 0.3455072463768116,
9
+ "std_success": 0.017157853434143458,
10
+ "mean_progress": 0.5624751563424671,
11
+ "mean_action_mse_to_best": 0.39550293604370906,
12
+ "gain_vs_h4": 0.04880724637681161,
13
+ "gain_vs_h16_rank_checkpoint": 0.04811594202898556,
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_scale0p25_safe_types_margin0p20.json",
18
+ "num_groups": 575,
19
+ "selection_mode": "retrieval_residual",
20
+ "num_candidates": 16,
21
+ "candidate_sigma": 0.0,
22
+ "selection_margin": 0.2,
23
+ "field_optim_steps": 0,
24
+ "field_optim_step_size": 0.0,
25
+ "field_optim_trust_radius": 0.0,
26
+ "field_optim_l2_penalty": 0.0,
27
+ "retrieval_neighbors": 1,
28
+ "retrieval_metric": "raw",
29
+ "retrieval_type_min_success": 0.0,
30
+ "retrieval_residual_scale": 0.25,
31
+ "retrieval_residual_anchor": "expert",
32
+ "policy_rollout_success_rate": 0.3373913043478261,
33
+ "policy_rollout_progress": 0.5497052084439961,
34
+ "oracle_success_rate": 0.8573913043478261,
35
+ "action_mse_to_best": 0.38161599399683915,
36
+ "best_policy_val": {
37
+ "bc_loss": 0.13721593966086706,
38
+ "field_effect_loss": 0.009290305380192068,
39
+ "field_potential_loss": 0.2666468388504452,
40
+ "field_preference_loss": 0.5130573478009965,
41
+ "lattice_edges": 3833.3333333333335,
42
+ "progress_mae": 0.1933159919248687,
43
+ "rank_acc": 0.8265031774838766,
44
+ "rank_loss": 0.5130523675017886,
45
+ "regret_mae": 0.3756548762321472,
46
+ "success_accuracy": 0.8773836526605818,
47
+ "total_loss": 1.5581054819954767
48
+ },
49
+ "per_task": {
50
+ "LiftPegUpright-v1": {
51
+ "action_mse_to_best": 0.31969772514973566,
52
+ "expert_success_rate": 0.8865979381443299,
53
+ "num_groups": 97,
54
+ "oracle_success_rate": 0.9278350515463918,
55
+ "policy_expert_regret": 1.0792337135248578,
56
+ "policy_oracle_regret": 1.1056491912333006,
57
+ "policy_rollout_progress": 0.570563180084081,
58
+ "policy_rollout_success_rate": 0.2268041237113402,
59
+ "restore_max_error": 4.76837158203125e-07
60
+ },
61
+ "PickCube-v1": {
62
+ "action_mse_to_best": 0.3026316432067408,
63
+ "expert_success_rate": 0.9375,
64
+ "num_groups": 208,
65
+ "oracle_success_rate": 0.9471153846153846,
66
+ "policy_expert_regret": 0.9610959745298785,
67
+ "policy_oracle_regret": 0.9745188419563839,
68
+ "policy_rollout_progress": 0.6161706692968997,
69
+ "policy_rollout_success_rate": 0.34134615384615385,
70
+ "restore_max_error": 4.76837158203125e-07
71
+ },
72
+ "PullCube-v1": {
73
+ "action_mse_to_best": 0.6063734511663388,
74
+ "expert_success_rate": 0.19480519480519481,
75
+ "num_groups": 77,
76
+ "oracle_success_rate": 0.36363636363636365,
77
+ "policy_expert_regret": 0.2203628039743461,
78
+ "policy_oracle_regret": 0.4144052575166978,
79
+ "policy_rollout_progress": 0.32507537823932414,
80
+ "policy_rollout_success_rate": 0.22077922077922077,
81
+ "restore_max_error": 4.76837158203125e-07
82
+ },
83
+ "PushCube-v1": {
84
+ "action_mse_to_best": 0.3676792739474393,
85
+ "expert_success_rate": 0.8617021276595744,
86
+ "num_groups": 94,
87
+ "oracle_success_rate": 0.9893617021276596,
88
+ "policy_expert_regret": 0.5719942746961371,
89
+ "policy_oracle_regret": 0.66845020953011,
90
+ "policy_rollout_progress": 0.682613620257124,
91
+ "policy_rollout_success_rate": 0.648936170212766,
92
+ "restore_max_error": 4.76837158203125e-07
93
+ },
94
+ "StackCube-v1": {
95
+ "action_mse_to_best": 0.4466517972806939,
96
+ "expert_success_rate": 0.6767676767676768,
97
+ "num_groups": 99,
98
+ "oracle_success_rate": 0.8585858585858586,
99
+ "policy_expert_regret": 0.97628669034351,
100
+ "policy_oracle_regret": 1.150446284901012,
101
+ "policy_rollout_progress": 0.43814022974534467,
102
+ "policy_rollout_success_rate": 0.23232323232323232,
103
+ "restore_max_error": 3.948807716369629e-07
104
+ }
105
+ }
106
+ },
107
+ {
108
+ "seed": 1,
109
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/near_miss_policy_bc5/seed_1/policy_rollout_retrieval_residual_scale0p25_safe_types_margin0p20.json",
110
+ "num_groups": 575,
111
+ "selection_mode": "retrieval_residual",
112
+ "num_candidates": 16,
113
+ "candidate_sigma": 0.0,
114
+ "selection_margin": 0.2,
115
+ "field_optim_steps": 0,
116
+ "field_optim_step_size": 0.0,
117
+ "field_optim_trust_radius": 0.0,
118
+ "field_optim_l2_penalty": 0.0,
119
+ "retrieval_neighbors": 1,
120
+ "retrieval_metric": "raw",
121
+ "retrieval_type_min_success": 0.0,
122
+ "retrieval_residual_scale": 0.25,
123
+ "retrieval_residual_anchor": "expert",
124
+ "policy_rollout_success_rate": 0.3339130434782609,
125
+ "policy_rollout_progress": 0.5583255888035764,
126
+ "oracle_success_rate": 0.8695652173913043,
127
+ "action_mse_to_best": 0.3887117049259984,
128
+ "best_policy_val": {
129
+ "bc_loss": 0.1315989200439718,
130
+ "field_effect_loss": 0.010110370814800262,
131
+ "field_potential_loss": 0.31546536915832096,
132
+ "field_preference_loss": 0.4960637440284093,
133
+ "lattice_edges": 3833.3333333333335,
134
+ "progress_mae": 0.19023226698239645,
135
+ "rank_acc": 0.8255469501018524,
136
+ "rank_loss": 0.4960531195004781,
137
+ "regret_mae": 0.42156532737943864,
138
+ "success_accuracy": 0.8807473679383596,
139
+ "total_loss": 1.561984618504842
140
+ },
141
+ "per_task": {
142
+ "LiftPegUpright-v1": {
143
+ "action_mse_to_best": 0.3259265956806794,
144
+ "expert_success_rate": 0.8584070796460177,
145
+ "num_groups": 113,
146
+ "oracle_success_rate": 0.9380530973451328,
147
+ "policy_expert_regret": 0.987694250535121,
148
+ "policy_oracle_regret": 1.066896405631462,
149
+ "policy_rollout_progress": 0.6352153201546289,
150
+ "policy_rollout_success_rate": 0.23008849557522124,
151
+ "restore_max_error": 4.76837158203125e-07
152
+ },
153
+ "PickCube-v1": {
154
+ "action_mse_to_best": 0.3195883755815094,
155
+ "expert_success_rate": 0.9402173913043478,
156
+ "num_groups": 184,
157
+ "oracle_success_rate": 0.9456521739130435,
158
+ "policy_expert_regret": 1.0787401368497345,
159
+ "policy_oracle_regret": 1.0787641413246647,
160
+ "policy_rollout_progress": 0.571896494983736,
161
+ "policy_rollout_success_rate": 0.28804347826086957,
162
+ "restore_max_error": 4.76837158203125e-07
163
+ },
164
+ "PullCube-v1": {
165
+ "action_mse_to_best": 0.5722180314754185,
166
+ "expert_success_rate": 0.25,
167
+ "num_groups": 76,
168
+ "oracle_success_rate": 0.40789473684210525,
169
+ "policy_expert_regret": 0.32544661396624225,
170
+ "policy_oracle_regret": 0.5978393972802319,
171
+ "policy_rollout_progress": 0.2542194458901098,
172
+ "policy_rollout_success_rate": 0.14473684210526316,
173
+ "restore_max_error": 4.76837158203125e-07
174
+ },
175
+ "PushCube-v1": {
176
+ "action_mse_to_best": 0.35717259357507164,
177
+ "expert_success_rate": 0.8198198198198198,
178
+ "num_groups": 111,
179
+ "oracle_success_rate": 1.0,
180
+ "policy_expert_regret": 0.3483100165654947,
181
+ "policy_oracle_regret": 0.4142704948380187,
182
+ "policy_rollout_progress": 0.8019457213781975,
183
+ "policy_rollout_success_rate": 0.7837837837837838,
184
+ "restore_max_error": 4.76837158203125e-07
185
+ },
186
+ "StackCube-v1": {
187
+ "action_mse_to_best": 0.49165423774252554,
188
+ "expert_success_rate": 0.6923076923076923,
189
+ "num_groups": 91,
190
+ "oracle_success_rate": 0.8571428571428571,
191
+ "policy_expert_regret": 1.128107481277906,
192
+ "policy_oracle_regret": 1.28739352114908,
193
+ "policy_rollout_progress": 0.39222279502140295,
194
+ "policy_rollout_success_rate": 0.16483516483516483,
195
+ "restore_max_error": 3.948807716369629e-07
196
+ }
197
+ }
198
+ },
199
+ {
200
+ "seed": 2,
201
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/near_miss_policy_bc5/seed_2/policy_rollout_retrieval_residual_scale0p25_safe_types_margin0p20.json",
202
+ "num_groups": 575,
203
+ "selection_mode": "retrieval_residual",
204
+ "num_candidates": 16,
205
+ "candidate_sigma": 0.0,
206
+ "selection_margin": 0.2,
207
+ "field_optim_steps": 0,
208
+ "field_optim_step_size": 0.0,
209
+ "field_optim_trust_radius": 0.0,
210
+ "field_optim_l2_penalty": 0.0,
211
+ "retrieval_neighbors": 1,
212
+ "retrieval_metric": "raw",
213
+ "retrieval_type_min_success": 0.0,
214
+ "retrieval_residual_scale": 0.25,
215
+ "retrieval_residual_anchor": "expert",
216
+ "policy_rollout_success_rate": 0.3652173913043478,
217
+ "policy_rollout_progress": 0.5793946717798288,
218
+ "oracle_success_rate": 0.8765217391304347,
219
+ "action_mse_to_best": 0.4161811092082897,
220
+ "best_policy_val": {
221
+ "bc_loss": 0.11367896075050037,
222
+ "field_effect_loss": 0.009670218582161598,
223
+ "field_potential_loss": 0.2641640139950646,
224
+ "field_preference_loss": 0.5130490180518892,
225
+ "lattice_edges": 3833.3333333333335,
226
+ "progress_mae": 0.2021729110015763,
227
+ "rank_acc": 0.8333857821093665,
228
+ "rank_loss": 0.5130119257503085,
229
+ "regret_mae": 0.3958987047274907,
230
+ "success_accuracy": 0.8680730561415354,
231
+ "total_loss": 1.4394984311527677
232
+ },
233
+ "per_task": {
234
+ "LiftPegUpright-v1": {
235
+ "action_mse_to_best": 0.35361564260771655,
236
+ "expert_success_rate": 0.8229166666666666,
237
+ "num_groups": 96,
238
+ "oracle_success_rate": 0.9270833333333334,
239
+ "policy_expert_regret": 0.8470402403424183,
240
+ "policy_oracle_regret": 0.9533149556567272,
241
+ "policy_rollout_progress": 0.6469602342694998,
242
+ "policy_rollout_success_rate": 0.3333333333333333,
243
+ "restore_max_error": 3.5762786865234375e-07
244
+ },
245
+ "PickCube-v1": {
246
+ "action_mse_to_best": 0.3353611076006083,
247
+ "expert_success_rate": 0.9444444444444444,
248
+ "num_groups": 198,
249
+ "oracle_success_rate": 0.9595959595959596,
250
+ "policy_expert_regret": 0.9954897118826406,
251
+ "policy_oracle_regret": 1.002573162411349,
252
+ "policy_rollout_progress": 0.6289936729114164,
253
+ "policy_rollout_success_rate": 0.32323232323232326,
254
+ "restore_max_error": 4.76837158203125e-07
255
+ },
256
+ "PullCube-v1": {
257
+ "action_mse_to_best": 0.6361265104677942,
258
+ "expert_success_rate": 0.24444444444444444,
259
+ "num_groups": 90,
260
+ "oracle_success_rate": 0.4666666666666667,
261
+ "policy_expert_regret": 0.29700109129425983,
262
+ "policy_oracle_regret": 0.4846563896440178,
263
+ "policy_rollout_progress": 0.3355572824607912,
264
+ "policy_rollout_success_rate": 0.23333333333333334,
265
+ "restore_max_error": 4.0978193283081055e-07
266
+ },
267
+ "PushCube-v1": {
268
+ "action_mse_to_best": 0.37776060026175906,
269
+ "expert_success_rate": 0.8514851485148515,
270
+ "num_groups": 101,
271
+ "oracle_success_rate": 1.0,
272
+ "policy_expert_regret": 0.36244124557712293,
273
+ "policy_oracle_regret": 0.43509086774717465,
274
+ "policy_rollout_progress": 0.792631904530053,
275
+ "policy_rollout_success_rate": 0.7722772277227723,
276
+ "restore_max_error": 4.76837158203125e-07
277
+ },
278
+ "StackCube-v1": {
279
+ "action_mse_to_best": 0.4838925581218468,
280
+ "expert_success_rate": 0.7666666666666667,
281
+ "num_groups": 90,
282
+ "oracle_success_rate": 0.9111111111111111,
283
+ "policy_expert_regret": 1.1635313358571795,
284
+ "policy_oracle_regret": 1.3126035009821255,
285
+ "policy_rollout_progress": 0.4027447640895844,
286
+ "policy_rollout_success_rate": 0.16666666666666666,
287
+ "restore_max_error": 4.76837158203125e-07
288
+ }
289
+ }
290
+ }
291
+ ]
292
+ }
results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_scale0p25_safe_types_margin0p20_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_scale0p25_safe_types_margin0p20.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: 34.55% +/- 1.72%
11
+ Gain vs h=16 rank checkpoint: +4.81%
12
+ Mean progress: 56.25%
13
+ Mean action MSE to best: 0.396
14
+
15
+ | seed | mode | k | retrieval K | retrieval metric | residual anchor | min type success | residual scale | margin | sigma | opt steps | trust | success | progress | oracle | action MSE |
16
+ |---:|---|---:|---:|---|---|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|
17
+ | 0 | retrieval_residual | 16 | 1 | raw | expert | 0.00 | 0.25 | 0.200 | 0.00 | 0 | 0.00 | 33.74% | 54.97% | 85.74% | 0.382 |
18
+ | 1 | retrieval_residual | 16 | 1 | raw | expert | 0.00 | 0.25 | 0.200 | 0.00 | 0 | 0.00 | 33.39% | 55.83% | 86.96% | 0.389 |
19
+ | 2 | retrieval_residual | 16 | 1 | raw | expert | 0.00 | 0.25 | 0.200 | 0.00 | 0 | 0.00 | 36.52% | 57.94% | 87.65% | 0.416 |
results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_scale0p40_safe_types_margin0p20_summary.json ADDED
@@ -0,0 +1,292 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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_scale0p40_safe_types_margin0p20.json",
5
+ "num_completed": 3,
6
+ "baseline_h4_policy_success": 0.2967,
7
+ "baseline_h16_rank_checkpoint_success": 0.29739130434782607,
8
+ "mean_success": 0.34782608695652173,
9
+ "std_success": 0.016771566540857326,
10
+ "mean_progress": 0.5654618425721729,
11
+ "mean_action_mse_to_best": 0.3968654996070309,
12
+ "gain_vs_h4": 0.05112608695652171,
13
+ "gain_vs_h16_rank_checkpoint": 0.05043478260869566,
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_scale0p40_safe_types_margin0p20.json",
18
+ "num_groups": 575,
19
+ "selection_mode": "retrieval_residual",
20
+ "num_candidates": 16,
21
+ "candidate_sigma": 0.0,
22
+ "selection_margin": 0.2,
23
+ "field_optim_steps": 0,
24
+ "field_optim_step_size": 0.0,
25
+ "field_optim_trust_radius": 0.0,
26
+ "field_optim_l2_penalty": 0.0,
27
+ "retrieval_neighbors": 1,
28
+ "retrieval_metric": "raw",
29
+ "retrieval_type_min_success": 0.0,
30
+ "retrieval_residual_scale": 0.4,
31
+ "retrieval_residual_anchor": "expert",
32
+ "policy_rollout_success_rate": 0.3408695652173913,
33
+ "policy_rollout_progress": 0.5532612886615907,
34
+ "oracle_success_rate": 0.8573913043478261,
35
+ "action_mse_to_best": 0.3826324706872844,
36
+ "best_policy_val": {
37
+ "bc_loss": 0.13721593966086706,
38
+ "field_effect_loss": 0.009290305380192068,
39
+ "field_potential_loss": 0.2666468388504452,
40
+ "field_preference_loss": 0.5130573478009965,
41
+ "lattice_edges": 3833.3333333333335,
42
+ "progress_mae": 0.1933159919248687,
43
+ "rank_acc": 0.8265031774838766,
44
+ "rank_loss": 0.5130523675017886,
45
+ "regret_mae": 0.3756548762321472,
46
+ "success_accuracy": 0.8773836526605818,
47
+ "total_loss": 1.5581054819954767
48
+ },
49
+ "per_task": {
50
+ "LiftPegUpright-v1": {
51
+ "action_mse_to_best": 0.31969772514973566,
52
+ "expert_success_rate": 0.8865979381443299,
53
+ "num_groups": 97,
54
+ "oracle_success_rate": 0.9278350515463918,
55
+ "policy_expert_regret": 1.0792337135248578,
56
+ "policy_oracle_regret": 1.1056491912333006,
57
+ "policy_rollout_progress": 0.570563180084081,
58
+ "policy_rollout_success_rate": 0.2268041237113402,
59
+ "restore_max_error": 4.76837158203125e-07
60
+ },
61
+ "PickCube-v1": {
62
+ "action_mse_to_best": 0.3028315781400754,
63
+ "expert_success_rate": 0.9375,
64
+ "num_groups": 208,
65
+ "oracle_success_rate": 0.9471153846153846,
66
+ "policy_expert_regret": 0.9612192197416264,
67
+ "policy_oracle_regret": 0.9746420871681318,
68
+ "policy_rollout_progress": 0.6160474240851517,
69
+ "policy_rollout_success_rate": 0.34134615384615385,
70
+ "restore_max_error": 4.76837158203125e-07
71
+ },
72
+ "PullCube-v1": {
73
+ "action_mse_to_best": 0.5994591196249057,
74
+ "expert_success_rate": 0.19480519480519481,
75
+ "num_groups": 77,
76
+ "oracle_success_rate": 0.36363636363636365,
77
+ "policy_expert_regret": 0.19730841831981458,
78
+ "policy_oracle_regret": 0.3691151456744108,
79
+ "policy_rollout_progress": 0.3443914641075852,
80
+ "policy_rollout_success_rate": 0.24675324675324675,
81
+ "restore_max_error": 4.76837158203125e-07
82
+ },
83
+ "PushCube-v1": {
84
+ "action_mse_to_best": 0.3661688497646692,
85
+ "expert_success_rate": 0.8617021276595744,
86
+ "num_groups": 94,
87
+ "oracle_success_rate": 0.9893617021276596,
88
+ "policy_expert_regret": 0.5719942746961371,
89
+ "policy_oracle_regret": 0.6688640964158038,
90
+ "policy_rollout_progress": 0.6821997333714303,
91
+ "policy_rollout_success_rate": 0.648936170212766,
92
+ "restore_max_error": 4.76837158203125e-07
93
+ },
94
+ "StackCube-v1": {
95
+ "action_mse_to_best": 0.4589474642780995,
96
+ "expert_success_rate": 0.6767676767676768,
97
+ "num_groups": 99,
98
+ "oracle_success_rate": 0.8585858585858586,
99
+ "policy_expert_regret": 0.9700043884792713,
100
+ "policy_oracle_regret": 1.1441639830367734,
101
+ "policy_rollout_progress": 0.4444225316095834,
102
+ "policy_rollout_success_rate": 0.23232323232323232,
103
+ "restore_max_error": 3.948807716369629e-07
104
+ }
105
+ }
106
+ },
107
+ {
108
+ "seed": 1,
109
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/near_miss_policy_bc5/seed_1/policy_rollout_retrieval_residual_scale0p40_safe_types_margin0p20.json",
110
+ "num_groups": 575,
111
+ "selection_mode": "retrieval_residual",
112
+ "num_candidates": 16,
113
+ "candidate_sigma": 0.0,
114
+ "selection_margin": 0.2,
115
+ "field_optim_steps": 0,
116
+ "field_optim_step_size": 0.0,
117
+ "field_optim_trust_radius": 0.0,
118
+ "field_optim_l2_penalty": 0.0,
119
+ "retrieval_neighbors": 1,
120
+ "retrieval_metric": "raw",
121
+ "retrieval_type_min_success": 0.0,
122
+ "retrieval_residual_scale": 0.4,
123
+ "retrieval_residual_anchor": "expert",
124
+ "policy_rollout_success_rate": 0.3356521739130435,
125
+ "policy_rollout_progress": 0.5607172309024179,
126
+ "oracle_success_rate": 0.8695652173913043,
127
+ "action_mse_to_best": 0.3895034541416427,
128
+ "best_policy_val": {
129
+ "bc_loss": 0.1315989200439718,
130
+ "field_effect_loss": 0.010110370814800262,
131
+ "field_potential_loss": 0.31546536915832096,
132
+ "field_preference_loss": 0.4960637440284093,
133
+ "lattice_edges": 3833.3333333333335,
134
+ "progress_mae": 0.19023226698239645,
135
+ "rank_acc": 0.8255469501018524,
136
+ "rank_loss": 0.4960531195004781,
137
+ "regret_mae": 0.42156532737943864,
138
+ "success_accuracy": 0.8807473679383596,
139
+ "total_loss": 1.561984618504842
140
+ },
141
+ "per_task": {
142
+ "LiftPegUpright-v1": {
143
+ "action_mse_to_best": 0.32557108186482064,
144
+ "expert_success_rate": 0.8584070796460177,
145
+ "num_groups": 113,
146
+ "oracle_success_rate": 0.9380530973451328,
147
+ "policy_expert_regret": 0.987694250535121,
148
+ "policy_oracle_regret": 1.066896405631462,
149
+ "policy_rollout_progress": 0.6352153201546289,
150
+ "policy_rollout_success_rate": 0.23008849557522124,
151
+ "restore_max_error": 4.76837158203125e-07
152
+ },
153
+ "PickCube-v1": {
154
+ "action_mse_to_best": 0.32029035203563777,
155
+ "expert_success_rate": 0.9402173913043478,
156
+ "num_groups": 184,
157
+ "oracle_success_rate": 0.9456521739130435,
158
+ "policy_expert_regret": 1.0790771580422702,
159
+ "policy_oracle_regret": 1.0791011625172005,
160
+ "policy_rollout_progress": 0.5715594737912002,
161
+ "policy_rollout_success_rate": 0.28804347826086957,
162
+ "restore_max_error": 4.76837158203125e-07
163
+ },
164
+ "PullCube-v1": {
165
+ "action_mse_to_best": 0.5839506075962594,
166
+ "expert_success_rate": 0.25,
167
+ "num_groups": 76,
168
+ "oracle_success_rate": 0.40789473684210525,
169
+ "policy_expert_regret": 0.2783891829465957,
170
+ "policy_oracle_regret": 0.5503785702595977,
171
+ "policy_rollout_progress": 0.275376009877379,
172
+ "policy_rollout_success_rate": 0.17105263157894737,
173
+ "restore_max_error": 4.76837158203125e-07
174
+ },
175
+ "PushCube-v1": {
176
+ "action_mse_to_best": 0.3557869063371474,
177
+ "expert_success_rate": 0.8198198198198198,
178
+ "num_groups": 111,
179
+ "oracle_success_rate": 1.0,
180
+ "policy_expert_regret": 0.3476062170020095,
181
+ "policy_oracle_regret": 0.4141964737896447,
182
+ "policy_rollout_progress": 0.8020197424265716,
183
+ "policy_rollout_success_rate": 0.7837837837837838,
184
+ "restore_max_error": 4.76837158203125e-07
185
+ },
186
+ "StackCube-v1": {
187
+ "action_mse_to_best": 0.4875707288508068,
188
+ "expert_success_rate": 0.6923076923076923,
189
+ "num_groups": 91,
190
+ "oracle_success_rate": 0.8571428571428571,
191
+ "policy_expert_regret": 1.1282008119991846,
192
+ "policy_oracle_regret": 1.3003485661286573,
193
+ "policy_rollout_progress": 0.3902567610308364,
194
+ "policy_rollout_success_rate": 0.15384615384615385,
195
+ "restore_max_error": 3.948807716369629e-07
196
+ }
197
+ }
198
+ },
199
+ {
200
+ "seed": 2,
201
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/near_miss_policy_bc5/seed_2/policy_rollout_retrieval_residual_scale0p40_safe_types_margin0p20.json",
202
+ "num_groups": 575,
203
+ "selection_mode": "retrieval_residual",
204
+ "num_candidates": 16,
205
+ "candidate_sigma": 0.0,
206
+ "selection_margin": 0.2,
207
+ "field_optim_steps": 0,
208
+ "field_optim_step_size": 0.0,
209
+ "field_optim_trust_radius": 0.0,
210
+ "field_optim_l2_penalty": 0.0,
211
+ "retrieval_neighbors": 1,
212
+ "retrieval_metric": "raw",
213
+ "retrieval_type_min_success": 0.0,
214
+ "retrieval_residual_scale": 0.4,
215
+ "retrieval_residual_anchor": "expert",
216
+ "policy_rollout_success_rate": 0.36695652173913046,
217
+ "policy_rollout_progress": 0.5824070081525101,
218
+ "oracle_success_rate": 0.8765217391304347,
219
+ "action_mse_to_best": 0.4184605739921655,
220
+ "best_policy_val": {
221
+ "bc_loss": 0.11367896075050037,
222
+ "field_effect_loss": 0.009670218582161598,
223
+ "field_potential_loss": 0.2641640139950646,
224
+ "field_preference_loss": 0.5130490180518892,
225
+ "lattice_edges": 3833.3333333333335,
226
+ "progress_mae": 0.2021729110015763,
227
+ "rank_acc": 0.8333857821093665,
228
+ "rank_loss": 0.5130119257503085,
229
+ "regret_mae": 0.3958987047274907,
230
+ "success_accuracy": 0.8680730561415354,
231
+ "total_loss": 1.4394984311527677
232
+ },
233
+ "per_task": {
234
+ "LiftPegUpright-v1": {
235
+ "action_mse_to_best": 0.3542889782741743,
236
+ "expert_success_rate": 0.8229166666666666,
237
+ "num_groups": 96,
238
+ "oracle_success_rate": 0.9270833333333334,
239
+ "policy_expert_regret": 0.8470402403424183,
240
+ "policy_oracle_regret": 0.9533149556567272,
241
+ "policy_rollout_progress": 0.6469602342694998,
242
+ "policy_rollout_success_rate": 0.3333333333333333,
243
+ "restore_max_error": 3.5762786865234375e-07
244
+ },
245
+ "PickCube-v1": {
246
+ "action_mse_to_best": 0.3368716507284629,
247
+ "expert_success_rate": 0.9444444444444444,
248
+ "num_groups": 198,
249
+ "oracle_success_rate": 0.9595959595959596,
250
+ "policy_expert_regret": 0.990470482309283,
251
+ "policy_oracle_regret": 0.9975539328379914,
252
+ "policy_rollout_progress": 0.628962397434269,
253
+ "policy_rollout_success_rate": 0.3282828282828283,
254
+ "restore_max_error": 4.76837158203125e-07
255
+ },
256
+ "PullCube-v1": {
257
+ "action_mse_to_best": 0.6334307440453105,
258
+ "expert_success_rate": 0.24444444444444444,
259
+ "num_groups": 90,
260
+ "oracle_success_rate": 0.4666666666666667,
261
+ "policy_expert_regret": 0.29369617260439956,
262
+ "policy_oracle_regret": 0.47999060957657397,
263
+ "policy_rollout_progress": 0.3483005565004836,
264
+ "policy_rollout_success_rate": 0.24444444444444444,
265
+ "restore_max_error": 4.0978193283081055e-07
266
+ },
267
+ "PushCube-v1": {
268
+ "action_mse_to_best": 0.38531619413654405,
269
+ "expert_success_rate": 0.8514851485148515,
270
+ "num_groups": 101,
271
+ "oracle_success_rate": 1.0,
272
+ "policy_expert_regret": 0.38044790141653306,
273
+ "policy_oracle_regret": 0.4353511377726451,
274
+ "policy_rollout_progress": 0.7923716345045826,
275
+ "policy_rollout_success_rate": 0.7722772277227723,
276
+ "restore_max_error": 4.76837158203125e-07
277
+ },
278
+ "StackCube-v1": {
279
+ "action_mse_to_best": 0.4886310968341099,
280
+ "expert_success_rate": 0.7666666666666667,
281
+ "num_groups": 90,
282
+ "oracle_success_rate": 0.9111111111111111,
283
+ "policy_expert_regret": 1.1548642312486967,
284
+ "policy_oracle_regret": 1.316851516895824,
285
+ "policy_rollout_progress": 0.40960785928699706,
286
+ "policy_rollout_success_rate": 0.15555555555555556,
287
+ "restore_max_error": 4.76837158203125e-07
288
+ }
289
+ }
290
+ }
291
+ ]
292
+ }
results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_scale0p40_safe_types_margin0p20_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_scale0p40_safe_types_margin0p20.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: 34.78% +/- 1.68%
11
+ Gain vs h=16 rank checkpoint: +5.04%
12
+ Mean progress: 56.55%
13
+ Mean action MSE to best: 0.397
14
+
15
+ | seed | mode | k | retrieval K | retrieval metric | residual anchor | min type success | residual scale | margin | sigma | opt steps | trust | success | progress | oracle | action MSE |
16
+ |---:|---|---:|---:|---|---|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|
17
+ | 0 | retrieval_residual | 16 | 1 | raw | expert | 0.00 | 0.40 | 0.200 | 0.00 | 0 | 0.00 | 34.09% | 55.33% | 85.74% | 0.383 |
18
+ | 1 | retrieval_residual | 16 | 1 | raw | expert | 0.00 | 0.40 | 0.200 | 0.00 | 0 | 0.00 | 33.57% | 56.07% | 86.96% | 0.390 |
19
+ | 2 | retrieval_residual | 16 | 1 | raw | expert | 0.00 | 0.40 | 0.200 | 0.00 | 0 | 0.00 | 36.70% | 58.24% | 87.65% | 0.418 |
results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_scale0p50_safe_types_margin0p20_summary.json ADDED
@@ -0,0 +1,292 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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_scale0p50_safe_types_margin0p20.json",
5
+ "num_completed": 3,
6
+ "baseline_h4_policy_success": 0.2967,
7
+ "baseline_h16_rank_checkpoint_success": 0.29739130434782607,
8
+ "mean_success": 0.34840579710144925,
9
+ "std_success": 0.017593032933905524,
10
+ "mean_progress": 0.5658890142956311,
11
+ "mean_action_mse_to_best": 0.3996537289498509,
12
+ "gain_vs_h4": 0.051705797101449236,
13
+ "gain_vs_h16_rank_checkpoint": 0.051014492753623186,
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_scale0p50_safe_types_margin0p20.json",
18
+ "num_groups": 575,
19
+ "selection_mode": "retrieval_residual",
20
+ "num_candidates": 16,
21
+ "candidate_sigma": 0.0,
22
+ "selection_margin": 0.2,
23
+ "field_optim_steps": 0,
24
+ "field_optim_step_size": 0.0,
25
+ "field_optim_trust_radius": 0.0,
26
+ "field_optim_l2_penalty": 0.0,
27
+ "retrieval_neighbors": 1,
28
+ "retrieval_metric": "raw",
29
+ "retrieval_type_min_success": 0.0,
30
+ "retrieval_residual_scale": 0.5,
31
+ "retrieval_residual_anchor": "expert",
32
+ "policy_rollout_success_rate": 0.3373913043478261,
33
+ "policy_rollout_progress": 0.5512449210683774,
34
+ "oracle_success_rate": 0.8573913043478261,
35
+ "action_mse_to_best": 0.3864748230957143,
36
+ "best_policy_val": {
37
+ "bc_loss": 0.13721593966086706,
38
+ "field_effect_loss": 0.009290305380192068,
39
+ "field_potential_loss": 0.2666468388504452,
40
+ "field_preference_loss": 0.5130573478009965,
41
+ "lattice_edges": 3833.3333333333335,
42
+ "progress_mae": 0.1933159919248687,
43
+ "rank_acc": 0.8265031774838766,
44
+ "rank_loss": 0.5130523675017886,
45
+ "regret_mae": 0.3756548762321472,
46
+ "success_accuracy": 0.8773836526605818,
47
+ "total_loss": 1.5581054819954767
48
+ },
49
+ "per_task": {
50
+ "LiftPegUpright-v1": {
51
+ "action_mse_to_best": 0.31969772514973566,
52
+ "expert_success_rate": 0.8865979381443299,
53
+ "num_groups": 97,
54
+ "oracle_success_rate": 0.9278350515463918,
55
+ "policy_expert_regret": 1.0792337135248578,
56
+ "policy_oracle_regret": 1.1056491912333006,
57
+ "policy_rollout_progress": 0.570563180084081,
58
+ "policy_rollout_success_rate": 0.2268041237113402,
59
+ "restore_max_error": 4.76837158203125e-07
60
+ },
61
+ "PickCube-v1": {
62
+ "action_mse_to_best": 0.3037318497394713,
63
+ "expert_success_rate": 0.9375,
64
+ "num_groups": 208,
65
+ "oracle_success_rate": 0.9471153846153846,
66
+ "policy_expert_regret": 0.9612761169600372,
67
+ "policy_oracle_regret": 0.9746989843865427,
68
+ "policy_rollout_progress": 0.6159905268667409,
69
+ "policy_rollout_success_rate": 0.34134615384615385,
70
+ "restore_max_error": 4.76837158203125e-07
71
+ },
72
+ "PullCube-v1": {
73
+ "action_mse_to_best": 0.6037939421155236,
74
+ "expert_success_rate": 0.19480519480519481,
75
+ "num_groups": 77,
76
+ "oracle_success_rate": 0.36363636363636365,
77
+ "policy_expert_regret": 0.19770652929308835,
78
+ "policy_oracle_regret": 0.41550173954252007,
79
+ "policy_rollout_progress": 0.3239788962135019,
80
+ "policy_rollout_success_rate": 0.22077922077922077,
81
+ "restore_max_error": 4.76837158203125e-07
82
+ },
83
+ "PushCube-v1": {
84
+ "action_mse_to_best": 0.36807565609032805,
85
+ "expert_success_rate": 0.8617021276595744,
86
+ "num_groups": 94,
87
+ "oracle_success_rate": 0.9893617021276596,
88
+ "policy_expert_regret": 0.5719942746961371,
89
+ "policy_oracle_regret": 0.6684110583143031,
90
+ "policy_rollout_progress": 0.6826527714729309,
91
+ "policy_rollout_success_rate": 0.648936170212766,
92
+ "restore_max_error": 4.76837158203125e-07
93
+ },
94
+ "StackCube-v1": {
95
+ "action_mse_to_best": 0.4741906462557087,
96
+ "expert_success_rate": 0.6767676767676768,
97
+ "num_groups": 99,
98
+ "oracle_success_rate": 0.8585858585858586,
99
+ "policy_expert_regret": 0.9661497884627545,
100
+ "policy_oracle_regret": 1.1403093830202564,
101
+ "policy_rollout_progress": 0.4482771316261003,
102
+ "policy_rollout_success_rate": 0.23232323232323232,
103
+ "restore_max_error": 3.948807716369629e-07
104
+ }
105
+ }
106
+ },
107
+ {
108
+ "seed": 1,
109
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/near_miss_policy_bc5/seed_1/policy_rollout_retrieval_residual_scale0p50_safe_types_margin0p20.json",
110
+ "num_groups": 575,
111
+ "selection_mode": "retrieval_residual",
112
+ "num_candidates": 16,
113
+ "candidate_sigma": 0.0,
114
+ "selection_margin": 0.2,
115
+ "field_optim_steps": 0,
116
+ "field_optim_step_size": 0.0,
117
+ "field_optim_trust_radius": 0.0,
118
+ "field_optim_l2_penalty": 0.0,
119
+ "retrieval_neighbors": 1,
120
+ "retrieval_metric": "raw",
121
+ "retrieval_type_min_success": 0.0,
122
+ "retrieval_residual_scale": 0.5,
123
+ "retrieval_residual_anchor": "expert",
124
+ "policy_rollout_success_rate": 0.3391304347826087,
125
+ "policy_rollout_progress": 0.5639093549484792,
126
+ "oracle_success_rate": 0.8695652173913043,
127
+ "action_mse_to_best": 0.391206984795306,
128
+ "best_policy_val": {
129
+ "bc_loss": 0.1315989200439718,
130
+ "field_effect_loss": 0.010110370814800262,
131
+ "field_potential_loss": 0.31546536915832096,
132
+ "field_preference_loss": 0.4960637440284093,
133
+ "lattice_edges": 3833.3333333333335,
134
+ "progress_mae": 0.19023226698239645,
135
+ "rank_acc": 0.8255469501018524,
136
+ "rank_loss": 0.4960531195004781,
137
+ "regret_mae": 0.42156532737943864,
138
+ "success_accuracy": 0.8807473679383596,
139
+ "total_loss": 1.561984618504842
140
+ },
141
+ "per_task": {
142
+ "LiftPegUpright-v1": {
143
+ "action_mse_to_best": 0.32557108186482064,
144
+ "expert_success_rate": 0.8584070796460177,
145
+ "num_groups": 113,
146
+ "oracle_success_rate": 0.9380530973451328,
147
+ "policy_expert_regret": 0.987694250535121,
148
+ "policy_oracle_regret": 1.066896405631462,
149
+ "policy_rollout_progress": 0.6352153201546289,
150
+ "policy_rollout_success_rate": 0.23008849557522124,
151
+ "restore_max_error": 4.76837158203125e-07
152
+ },
153
+ "PickCube-v1": {
154
+ "action_mse_to_best": 0.32013416324701643,
155
+ "expert_success_rate": 0.9402173913043478,
156
+ "num_groups": 184,
157
+ "oracle_success_rate": 0.9456521739130435,
158
+ "policy_expert_regret": 1.0789482782129198,
159
+ "policy_oracle_regret": 1.07897228268785,
160
+ "policy_rollout_progress": 0.5716883536205506,
161
+ "policy_rollout_success_rate": 0.28804347826086957,
162
+ "restore_max_error": 4.76837158203125e-07
163
+ },
164
+ "PullCube-v1": {
165
+ "action_mse_to_best": 0.5884782922895331,
166
+ "expert_success_rate": 0.25,
167
+ "num_groups": 76,
168
+ "oracle_success_rate": 0.40789473684210525,
169
+ "policy_expert_regret": 0.30238124975738556,
170
+ "policy_oracle_regret": 0.5492499411155126,
171
+ "policy_rollout_progress": 0.2765046390214641,
172
+ "policy_rollout_success_rate": 0.17105263157894737,
173
+ "restore_max_error": 4.76837158203125e-07
174
+ },
175
+ "PushCube-v1": {
176
+ "action_mse_to_best": 0.36165393194234047,
177
+ "expert_success_rate": 0.8198198198198198,
178
+ "num_groups": 111,
179
+ "oracle_success_rate": 1.0,
180
+ "policy_expert_regret": 0.33009791642695935,
181
+ "policy_oracle_regret": 0.3798210240967639,
182
+ "policy_rollout_progress": 0.8183771741014343,
183
+ "policy_rollout_success_rate": 0.8018018018018018,
184
+ "restore_max_error": 4.76837158203125e-07
185
+ },
186
+ "StackCube-v1": {
187
+ "action_mse_to_best": 0.4877127616430868,
188
+ "expert_success_rate": 0.6923076923076923,
189
+ "num_groups": 91,
190
+ "oracle_success_rate": 0.8571428571428571,
191
+ "policy_expert_regret": 1.1291864525813322,
192
+ "policy_oracle_regret": 1.301334206710805,
193
+ "policy_rollout_progress": 0.3892711204486889,
194
+ "policy_rollout_success_rate": 0.15384615384615385,
195
+ "restore_max_error": 3.948807716369629e-07
196
+ }
197
+ }
198
+ },
199
+ {
200
+ "seed": 2,
201
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/near_miss_policy_bc5/seed_2/policy_rollout_retrieval_residual_scale0p50_safe_types_margin0p20.json",
202
+ "num_groups": 575,
203
+ "selection_mode": "retrieval_residual",
204
+ "num_candidates": 16,
205
+ "candidate_sigma": 0.0,
206
+ "selection_margin": 0.2,
207
+ "field_optim_steps": 0,
208
+ "field_optim_step_size": 0.0,
209
+ "field_optim_trust_radius": 0.0,
210
+ "field_optim_l2_penalty": 0.0,
211
+ "retrieval_neighbors": 1,
212
+ "retrieval_metric": "raw",
213
+ "retrieval_type_min_success": 0.0,
214
+ "retrieval_residual_scale": 0.5,
215
+ "retrieval_residual_anchor": "expert",
216
+ "policy_rollout_success_rate": 0.36869565217391304,
217
+ "policy_rollout_progress": 0.5825127668700367,
218
+ "oracle_success_rate": 0.8765217391304347,
219
+ "action_mse_to_best": 0.42127937895853235,
220
+ "best_policy_val": {
221
+ "bc_loss": 0.11367896075050037,
222
+ "field_effect_loss": 0.009670218582161598,
223
+ "field_potential_loss": 0.2641640139950646,
224
+ "field_preference_loss": 0.5130490180518892,
225
+ "lattice_edges": 3833.3333333333335,
226
+ "progress_mae": 0.2021729110015763,
227
+ "rank_acc": 0.8333857821093665,
228
+ "rank_loss": 0.5130119257503085,
229
+ "regret_mae": 0.3958987047274907,
230
+ "success_accuracy": 0.8680730561415354,
231
+ "total_loss": 1.4394984311527677
232
+ },
233
+ "per_task": {
234
+ "LiftPegUpright-v1": {
235
+ "action_mse_to_best": 0.3546536646366197,
236
+ "expert_success_rate": 0.8229166666666666,
237
+ "num_groups": 96,
238
+ "oracle_success_rate": 0.9270833333333334,
239
+ "policy_expert_regret": 0.8470402403424183,
240
+ "policy_oracle_regret": 0.9533149556567272,
241
+ "policy_rollout_progress": 0.6469602342694998,
242
+ "policy_rollout_success_rate": 0.3333333333333333,
243
+ "restore_max_error": 3.5762786865234375e-07
244
+ },
245
+ "PickCube-v1": {
246
+ "action_mse_to_best": 0.3389337710245992,
247
+ "expert_success_rate": 0.9444444444444444,
248
+ "num_groups": 198,
249
+ "oracle_success_rate": 0.9595959595959596,
250
+ "policy_expert_regret": 0.9908797259000365,
251
+ "policy_oracle_regret": 0.9979631764287449,
252
+ "policy_rollout_progress": 0.6285531538435154,
253
+ "policy_rollout_success_rate": 0.3282828282828283,
254
+ "restore_max_error": 4.76837158203125e-07
255
+ },
256
+ "PullCube-v1": {
257
+ "action_mse_to_best": 0.6297647087110414,
258
+ "expert_success_rate": 0.24444444444444444,
259
+ "num_groups": 90,
260
+ "oracle_success_rate": 0.4666666666666667,
261
+ "policy_expert_regret": 0.2943164942855017,
262
+ "policy_oracle_regret": 0.5019078844192336,
263
+ "policy_rollout_progress": 0.33749439276893506,
264
+ "policy_rollout_success_rate": 0.23333333333333334,
265
+ "restore_max_error": 4.0978193283081055e-07
266
+ },
267
+ "PushCube-v1": {
268
+ "action_mse_to_best": 0.3902471246990827,
269
+ "expert_success_rate": 0.8514851485148515,
270
+ "num_groups": 101,
271
+ "oracle_success_rate": 1.0,
272
+ "policy_expert_regret": 0.38038931005071885,
273
+ "policy_oracle_regret": 0.4174000021844807,
274
+ "policy_rollout_progress": 0.8004217799937371,
275
+ "policy_rollout_success_rate": 0.7821782178217822,
276
+ "restore_max_error": 4.76837158203125e-07
277
+ },
278
+ "StackCube-v1": {
279
+ "action_mse_to_best": 0.499846900606321,
280
+ "expert_success_rate": 0.7666666666666667,
281
+ "num_groups": 90,
282
+ "oracle_success_rate": 0.9111111111111111,
283
+ "policy_expert_regret": 1.153320112493303,
284
+ "policy_oracle_regret": 1.3023922776182493,
285
+ "policy_rollout_progress": 0.4129559874534607,
286
+ "policy_rollout_success_rate": 0.16666666666666666,
287
+ "restore_max_error": 4.76837158203125e-07
288
+ }
289
+ }
290
+ }
291
+ ]
292
+ }
results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_scale0p50_safe_types_margin0p20_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_scale0p50_safe_types_margin0p20.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: 34.84% +/- 1.76%
11
+ Gain vs h=16 rank checkpoint: +5.10%
12
+ Mean progress: 56.59%
13
+ Mean action MSE to best: 0.400
14
+
15
+ | seed | mode | k | retrieval K | retrieval metric | residual anchor | min type success | residual scale | margin | sigma | opt steps | trust | success | progress | oracle | action MSE |
16
+ |---:|---|---:|---:|---|---|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|
17
+ | 0 | retrieval_residual | 16 | 1 | raw | expert | 0.00 | 0.50 | 0.200 | 0.00 | 0 | 0.00 | 33.74% | 55.12% | 85.74% | 0.386 |
18
+ | 1 | retrieval_residual | 16 | 1 | raw | expert | 0.00 | 0.50 | 0.200 | 0.00 | 0 | 0.00 | 33.91% | 56.39% | 86.96% | 0.391 |
19
+ | 2 | retrieval_residual | 16 | 1 | raw | expert | 0.00 | 0.50 | 0.200 | 0.00 | 0 | 0.00 | 36.87% | 58.25% | 87.65% | 0.421 |
results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_scale0p75_safe_types_margin0p20_summary.json ADDED
@@ -0,0 +1,292 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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_scale0p75_safe_types_margin0p20.json",
5
+ "num_completed": 3,
6
+ "baseline_h4_policy_success": 0.2967,
7
+ "baseline_h16_rank_checkpoint_success": 0.29739130434782607,
8
+ "mean_success": 0.3443478260869565,
9
+ "std_success": 0.013803919883815241,
10
+ "mean_progress": 0.5630646219364133,
11
+ "mean_action_mse_to_best": 0.43697492087772793,
12
+ "gain_vs_h4": 0.0476478260869565,
13
+ "gain_vs_h16_rank_checkpoint": 0.04695652173913045,
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_scale0p75_safe_types_margin0p20.json",
18
+ "num_groups": 575,
19
+ "selection_mode": "retrieval_residual",
20
+ "num_candidates": 16,
21
+ "candidate_sigma": 0.0,
22
+ "selection_margin": 0.2,
23
+ "field_optim_steps": 0,
24
+ "field_optim_step_size": 0.0,
25
+ "field_optim_trust_radius": 0.0,
26
+ "field_optim_l2_penalty": 0.0,
27
+ "retrieval_neighbors": 1,
28
+ "retrieval_metric": "raw",
29
+ "retrieval_type_min_success": 0.0,
30
+ "retrieval_residual_scale": 0.75,
31
+ "retrieval_residual_anchor": "expert",
32
+ "policy_rollout_success_rate": 0.3339130434782609,
33
+ "policy_rollout_progress": 0.5473541649864258,
34
+ "oracle_success_rate": 0.8573913043478261,
35
+ "action_mse_to_best": 0.4557102193934438,
36
+ "best_policy_val": {
37
+ "bc_loss": 0.13721593966086706,
38
+ "field_effect_loss": 0.009290305380192068,
39
+ "field_potential_loss": 0.2666468388504452,
40
+ "field_preference_loss": 0.5130573478009965,
41
+ "lattice_edges": 3833.3333333333335,
42
+ "progress_mae": 0.1933159919248687,
43
+ "rank_acc": 0.8265031774838766,
44
+ "rank_loss": 0.5130523675017886,
45
+ "regret_mae": 0.3756548762321472,
46
+ "success_accuracy": 0.8773836526605818,
47
+ "total_loss": 1.5581054819954767
48
+ },
49
+ "per_task": {
50
+ "LiftPegUpright-v1": {
51
+ "action_mse_to_best": 0.5816490878255021,
52
+ "expert_success_rate": 0.8865979381443299,
53
+ "num_groups": 97,
54
+ "oracle_success_rate": 0.9278350515463918,
55
+ "policy_expert_regret": 1.1229144115116179,
56
+ "policy_oracle_regret": 1.1493298892200607,
57
+ "policy_rollout_progress": 0.5371917604478365,
58
+ "policy_rollout_success_rate": 0.21649484536082475,
59
+ "restore_max_error": 4.76837158203125e-07
60
+ },
61
+ "PickCube-v1": {
62
+ "action_mse_to_best": 0.3150205578511724,
63
+ "expert_success_rate": 0.9375,
64
+ "num_groups": 208,
65
+ "oracle_success_rate": 0.9471153846153846,
66
+ "policy_expert_regret": 0.9613065121767039,
67
+ "policy_oracle_regret": 0.9747293796032094,
68
+ "policy_rollout_progress": 0.6159601316500741,
69
+ "policy_rollout_success_rate": 0.34134615384615385,
70
+ "restore_max_error": 4.76837158203125e-07
71
+ },
72
+ "PullCube-v1": {
73
+ "action_mse_to_best": 0.6096085838296197,
74
+ "expert_success_rate": 0.19480519480519481,
75
+ "num_groups": 77,
76
+ "oracle_success_rate": 0.36363636363636365,
77
+ "policy_expert_regret": 0.220309721780699,
78
+ "policy_oracle_regret": 0.43741786793515375,
79
+ "policy_rollout_progress": 0.3150497808078812,
80
+ "policy_rollout_success_rate": 0.2077922077922078,
81
+ "restore_max_error": 4.76837158203125e-07
82
+ },
83
+ "PushCube-v1": {
84
+ "action_mse_to_best": 0.3728606113965841,
85
+ "expert_success_rate": 0.8617021276595744,
86
+ "num_groups": 94,
87
+ "oracle_success_rate": 0.9893617021276596,
88
+ "policy_expert_regret": 0.5719942746961371,
89
+ "policy_oracle_regret": 0.6495450359392674,
90
+ "policy_rollout_progress": 0.6908804959756263,
91
+ "policy_rollout_success_rate": 0.6595744680851063,
92
+ "restore_max_error": 4.76837158203125e-07
93
+ },
94
+ "StackCube-v1": {
95
+ "action_mse_to_best": 0.5868725269995253,
96
+ "expert_success_rate": 0.6767676767676768,
97
+ "num_groups": 99,
98
+ "oracle_success_rate": 0.8585858585858586,
99
+ "policy_expert_regret": 0.9729163908597195,
100
+ "policy_oracle_regret": 1.1433258466046266,
101
+ "policy_rollout_progress": 0.4575731007739751,
102
+ "policy_rollout_success_rate": 0.2222222222222222,
103
+ "restore_max_error": 3.948807716369629e-07
104
+ }
105
+ }
106
+ },
107
+ {
108
+ "seed": 1,
109
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/near_miss_policy_bc5/seed_1/policy_rollout_retrieval_residual_scale0p75_safe_types_margin0p20.json",
110
+ "num_groups": 575,
111
+ "selection_mode": "retrieval_residual",
112
+ "num_candidates": 16,
113
+ "candidate_sigma": 0.0,
114
+ "selection_margin": 0.2,
115
+ "field_optim_steps": 0,
116
+ "field_optim_step_size": 0.0,
117
+ "field_optim_trust_radius": 0.0,
118
+ "field_optim_l2_penalty": 0.0,
119
+ "retrieval_neighbors": 1,
120
+ "retrieval_metric": "raw",
121
+ "retrieval_type_min_success": 0.0,
122
+ "retrieval_residual_scale": 0.75,
123
+ "retrieval_residual_anchor": "expert",
124
+ "policy_rollout_success_rate": 0.3391304347826087,
125
+ "policy_rollout_progress": 0.5640019817455956,
126
+ "oracle_success_rate": 0.8695652173913043,
127
+ "action_mse_to_best": 0.40411116663852464,
128
+ "best_policy_val": {
129
+ "bc_loss": 0.1315989200439718,
130
+ "field_effect_loss": 0.010110370814800262,
131
+ "field_potential_loss": 0.31546536915832096,
132
+ "field_preference_loss": 0.4960637440284093,
133
+ "lattice_edges": 3833.3333333333335,
134
+ "progress_mae": 0.19023226698239645,
135
+ "rank_acc": 0.8255469501018524,
136
+ "rank_loss": 0.4960531195004781,
137
+ "regret_mae": 0.42156532737943864,
138
+ "success_accuracy": 0.8807473679383596,
139
+ "total_loss": 1.561984618504842
140
+ },
141
+ "per_task": {
142
+ "LiftPegUpright-v1": {
143
+ "action_mse_to_best": 0.3337573827681156,
144
+ "expert_success_rate": 0.8584070796460177,
145
+ "num_groups": 113,
146
+ "oracle_success_rate": 0.9380530973451328,
147
+ "policy_expert_regret": 0.987694250535121,
148
+ "policy_oracle_regret": 1.066896405631462,
149
+ "policy_rollout_progress": 0.6352153201546289,
150
+ "policy_rollout_success_rate": 0.23008849557522124,
151
+ "restore_max_error": 4.76837158203125e-07
152
+ },
153
+ "PickCube-v1": {
154
+ "action_mse_to_best": 0.32512720481937996,
155
+ "expert_success_rate": 0.9402173913043478,
156
+ "num_groups": 184,
157
+ "oracle_success_rate": 0.9456521739130435,
158
+ "policy_expert_regret": 1.0788817837955835,
159
+ "policy_oracle_regret": 1.0789057882705138,
160
+ "policy_rollout_progress": 0.5717548480378869,
161
+ "policy_rollout_success_rate": 0.28804347826086957,
162
+ "restore_max_error": 4.76837158203125e-07
163
+ },
164
+ "PullCube-v1": {
165
+ "action_mse_to_best": 0.5897331627968111,
166
+ "expert_success_rate": 0.25,
167
+ "num_groups": 76,
168
+ "oracle_success_rate": 0.40789473684210525,
169
+ "policy_expert_regret": 0.30176124922735126,
170
+ "policy_oracle_regret": 0.5468284703889176,
171
+ "policy_rollout_progress": 0.2789261097480592,
172
+ "policy_rollout_success_rate": 0.17105263157894737,
173
+ "restore_max_error": 4.76837158203125e-07
174
+ },
175
+ "PushCube-v1": {
176
+ "action_mse_to_best": 0.37386467022536035,
177
+ "expert_success_rate": 0.8198198198198198,
178
+ "num_groups": 111,
179
+ "oracle_success_rate": 1.0,
180
+ "policy_expert_regret": 0.34603070796609997,
181
+ "policy_oracle_regret": 0.3794788784003473,
182
+ "policy_rollout_progress": 0.818719319797851,
183
+ "policy_rollout_success_rate": 0.8018018018018018,
184
+ "restore_max_error": 4.76837158203125e-07
185
+ },
186
+ "StackCube-v1": {
187
+ "action_mse_to_best": 0.5330465067034731,
188
+ "expert_success_rate": 0.6923076923076923,
189
+ "num_groups": 91,
190
+ "oracle_success_rate": 0.8571428571428571,
191
+ "policy_expert_regret": 1.1311752933722277,
192
+ "policy_oracle_regret": 1.3033230475017004,
193
+ "policy_rollout_progress": 0.38728227965779355,
194
+ "policy_rollout_success_rate": 0.15384615384615385,
195
+ "restore_max_error": 3.948807716369629e-07
196
+ }
197
+ }
198
+ },
199
+ {
200
+ "seed": 2,
201
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/near_miss_policy_bc5/seed_2/policy_rollout_retrieval_residual_scale0p75_safe_types_margin0p20.json",
202
+ "num_groups": 575,
203
+ "selection_mode": "retrieval_residual",
204
+ "num_candidates": 16,
205
+ "candidate_sigma": 0.0,
206
+ "selection_margin": 0.2,
207
+ "field_optim_steps": 0,
208
+ "field_optim_step_size": 0.0,
209
+ "field_optim_trust_radius": 0.0,
210
+ "field_optim_l2_penalty": 0.0,
211
+ "retrieval_neighbors": 1,
212
+ "retrieval_metric": "raw",
213
+ "retrieval_type_min_success": 0.0,
214
+ "retrieval_residual_scale": 0.75,
215
+ "retrieval_residual_anchor": "expert",
216
+ "policy_rollout_success_rate": 0.36,
217
+ "policy_rollout_progress": 0.5778377190772185,
218
+ "oracle_success_rate": 0.8765217391304347,
219
+ "action_mse_to_best": 0.4511033766012153,
220
+ "best_policy_val": {
221
+ "bc_loss": 0.11367896075050037,
222
+ "field_effect_loss": 0.009670218582161598,
223
+ "field_potential_loss": 0.2641640139950646,
224
+ "field_preference_loss": 0.5130490180518892,
225
+ "lattice_edges": 3833.3333333333335,
226
+ "progress_mae": 0.2021729110015763,
227
+ "rank_acc": 0.8333857821093665,
228
+ "rank_loss": 0.5130119257503085,
229
+ "regret_mae": 0.3958987047274907,
230
+ "success_accuracy": 0.8680730561415354,
231
+ "total_loss": 1.4394984311527677
232
+ },
233
+ "per_task": {
234
+ "LiftPegUpright-v1": {
235
+ "action_mse_to_best": 0.35361564260771655,
236
+ "expert_success_rate": 0.8229166666666666,
237
+ "num_groups": 96,
238
+ "oracle_success_rate": 0.9270833333333334,
239
+ "policy_expert_regret": 0.8470402403424183,
240
+ "policy_oracle_regret": 0.9533149556567272,
241
+ "policy_rollout_progress": 0.6469602342694998,
242
+ "policy_rollout_success_rate": 0.3333333333333333,
243
+ "restore_max_error": 3.5762786865234375e-07
244
+ },
245
+ "PickCube-v1": {
246
+ "action_mse_to_best": 0.3599281292408705,
247
+ "expert_success_rate": 0.9444444444444444,
248
+ "num_groups": 198,
249
+ "oracle_success_rate": 0.9595959595959596,
250
+ "policy_expert_regret": 1.0049364137736083,
251
+ "policy_oracle_regret": 1.0120198643023166,
252
+ "policy_rollout_progress": 0.6245974760709537,
253
+ "policy_rollout_success_rate": 0.3181818181818182,
254
+ "restore_max_error": 4.76837158203125e-07
255
+ },
256
+ "PullCube-v1": {
257
+ "action_mse_to_best": 0.6444657590654161,
258
+ "expert_success_rate": 0.24444444444444444,
259
+ "num_groups": 90,
260
+ "oracle_success_rate": 0.4666666666666667,
261
+ "policy_expert_regret": 0.2929868201276703,
262
+ "policy_oracle_regret": 0.5005048314799245,
263
+ "policy_rollout_progress": 0.33889744570824404,
264
+ "policy_rollout_success_rate": 0.23333333333333334,
265
+ "restore_max_error": 4.0978193283081055e-07
266
+ },
267
+ "PushCube-v1": {
268
+ "action_mse_to_best": 0.41782953700806835,
269
+ "expert_success_rate": 0.8514851485148515,
270
+ "num_groups": 101,
271
+ "oracle_success_rate": 1.0,
272
+ "policy_expert_regret": 0.4179433489790057,
273
+ "policy_oracle_regret": 0.4727004269267073,
274
+ "policy_rollout_progress": 0.7748243255485402,
275
+ "policy_rollout_success_rate": 0.7524752475247525,
276
+ "restore_max_error": 4.76837158203125e-07
277
+ },
278
+ "StackCube-v1": {
279
+ "action_mse_to_best": 0.5996540967995921,
280
+ "expert_success_rate": 0.7666666666666667,
281
+ "num_groups": 90,
282
+ "oracle_success_rate": 0.9111111111111111,
283
+ "policy_expert_regret": 1.150950428015656,
284
+ "policy_oracle_regret": 1.3000225931406022,
285
+ "policy_rollout_progress": 0.4191130969259474,
286
+ "policy_rollout_success_rate": 0.16666666666666666,
287
+ "restore_max_error": 4.76837158203125e-07
288
+ }
289
+ }
290
+ }
291
+ ]
292
+ }
results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_scale0p75_safe_types_margin0p20_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_scale0p75_safe_types_margin0p20.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: 34.43% +/- 1.38%
11
+ Gain vs h=16 rank checkpoint: +4.70%
12
+ Mean progress: 56.31%
13
+ Mean action MSE to best: 0.437
14
+
15
+ | seed | mode | k | retrieval K | retrieval metric | residual anchor | min type success | residual scale | margin | sigma | opt steps | trust | success | progress | oracle | action MSE |
16
+ |---:|---|---:|---:|---|---|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|
17
+ | 0 | retrieval_residual | 16 | 1 | raw | expert | 0.00 | 0.75 | 0.200 | 0.00 | 0 | 0.00 | 33.39% | 54.74% | 85.74% | 0.456 |
18
+ | 1 | retrieval_residual | 16 | 1 | raw | expert | 0.00 | 0.75 | 0.200 | 0.00 | 0 | 0.00 | 33.91% | 56.40% | 86.96% | 0.404 |
19
+ | 2 | retrieval_residual | 16 | 1 | raw | expert | 0.00 | 0.75 | 0.200 | 0.00 | 0 | 0.00 | 36.00% | 57.78% | 87.65% | 0.451 |