anhtld commited on
Commit
20f6dfd
·
verified ·
1 Parent(s): 8bc4ad0

Auto-sync: 2026-06-28 01:47:01 (part 2)

Browse files
results/README.md CHANGED
@@ -22,14 +22,21 @@ Use these summaries for the current paper tables:
22
  - `h16_policy_ckpt_near_miss_policy_bc5_bestpt_field_sweep_summary.md`
23
  - `h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_v2_summary.md`
24
  - `h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_knn4_v2_summary.md`
25
- - pending scale-sweep summaries:
26
  `h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_scale{0p25,0p50,0p75,1p25}_summary.md`
27
- - pending hybrid summaries:
 
 
 
 
 
28
  `h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_hybrid_k{32,64}_sigma*.md`
29
  - `h16_policy_ckpt_nonexpert_policy_bc5_summary.md`
30
  - `h16_policy_ckpt_nonexpert_policy_bc5_bestpt_field_sweep_summary.md`
31
  - `h16_policy_ckpt_field_selected_noexpert_bc5_summary.md`
32
  - `h16_policy_ckpt_field_selected_noexpert_bc5_bestpt_field_sweep_summary.md`
 
 
33
 
34
  Known stale or incomplete files:
35
 
@@ -63,35 +70,39 @@ Completed deployment-clean diagnostics:
63
  no-expert candidates for the rest.
64
  - Jobs `14858328`-`14858333` completed train-split field-teacher distillation:
65
  direct student rollout is 26.84%, and field-guided rollout is 27.65%.
 
 
 
66
 
67
- Active field-teacher distillation jobs:
68
 
69
- Last checked: `2026-06-28 05:26 UTC`.
70
 
71
- - Job `14858449` completed the all-split target map export for aligned validation
72
- checkpoint selection and seed-invariant student train coverage.
73
- - Job `14858450` completed 3 seeds of `field_selected_noexpert_bc5_allmap` with
74
- 100% train/val target-map coverage.
75
- - Jobs `14858451`/`14858452` and `14858453`/`14858454` are pending priority and
76
- will evaluate direct and field-guided allmap rollouts.
77
- - Job `14858455` rebuilds `paper_table_status.*` after the allmap summaries.
78
-
79
- Active residual transport jobs:
80
-
81
- - Code now supports `--retrieval-residual-scale`, defaulting to `1.0` so the
82
  completed 32.12% result is unchanged.
83
  - Job `14858978` completed the CPU Apptainer unit smoke for residual scale
84
  selection after smoke jobs `14858889`/`14858894` caught two wiring bugs before
85
  rollout jobs started.
86
- - Jobs `14858875`/`14858876`, `14858877`/`14858878`,
87
- `14858879`/`14858880`, and `14858881`/`14858882` evaluate residual scales
88
- `0.25`, `0.50`, `0.75`, and `1.25`.
89
- - Job `14858883` rebuilds `paper_table_status.*` after the scale summaries.
90
  - Job `14859041` completed the CPU Apptainer unit smoke for hybrid
91
  residual+Gaussian selection.
92
- - Jobs `14859042`/`14859043` and `14859044`/`14859045` evaluate residual+Gaussian
93
- hybrid proposals at K32/sigma0.35 and K64/sigma0.50.
94
- - Job `14859046` rebuilds `paper_table_status.*` after the hybrid summaries.
 
 
 
 
 
 
 
 
 
 
 
 
95
 
96
  Paper table automation:
97
 
 
22
  - `h16_policy_ckpt_near_miss_policy_bc5_bestpt_field_sweep_summary.md`
23
  - `h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_v2_summary.md`
24
  - `h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_knn4_v2_summary.md`
25
+ - residual scale-sweep summaries:
26
  `h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_scale{0p25,0p50,0p75,1p25}_summary.md`
27
+ - residual-family mask summaries, currently active:
28
+ `h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_scale0p50_no_random_summary.md`,
29
+ `h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_scale0p50_no_random_wrongdir_summary.md`,
30
+ `h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_scale0p25_no_random_wrongdir_summary.md`,
31
+ `h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_scale0p50_safe_types_summary.md`
32
+ - hybrid summaries:
33
  `h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_hybrid_k{32,64}_sigma*.md`
34
  - `h16_policy_ckpt_nonexpert_policy_bc5_summary.md`
35
  - `h16_policy_ckpt_nonexpert_policy_bc5_bestpt_field_sweep_summary.md`
36
  - `h16_policy_ckpt_field_selected_noexpert_bc5_summary.md`
37
  - `h16_policy_ckpt_field_selected_noexpert_bc5_bestpt_field_sweep_summary.md`
38
+ - `h16_policy_ckpt_field_selected_noexpert_bc5_allmap_summary.md`
39
+ - `h16_policy_ckpt_field_selected_noexpert_bc5_allmap_bestpt_field_sweep_summary.md`
40
 
41
  Known stale or incomplete files:
42
 
 
70
  no-expert candidates for the rest.
71
  - Jobs `14858328`-`14858333` completed train-split field-teacher distillation:
72
  direct student rollout is 26.84%, and field-guided rollout is 27.65%.
73
+ - Jobs `14858449`-`14858455` completed aligned all-split field-teacher
74
+ distillation: direct student rollout is 28.00%, and field-guided rollout is
75
+ 26.49%.
76
 
77
+ Completed residual transport jobs:
78
 
79
+ Last checked: `2026-06-28 05:47 UTC`.
80
 
81
+ - Code supports `--retrieval-residual-scale`, defaulting to `1.0` so the
 
 
 
 
 
 
 
 
 
 
82
  completed 32.12% result is unchanged.
83
  - Job `14858978` completed the CPU Apptainer unit smoke for residual scale
84
  selection after smoke jobs `14858889`/`14858894` caught two wiring bugs before
85
  rollout jobs started.
86
+ - Jobs `14858875`-`14858883` completed residual scales `0.25`, `0.50`, `0.75`,
87
+ and `1.25`. Scale `0.50` is the current best clean deployment bridge at
88
+ 33.33%; scale `0.25` ties the previous clean best at 32.93%.
 
89
  - Job `14859041` completed the CPU Apptainer unit smoke for hybrid
90
  residual+Gaussian selection.
91
+ - Jobs `14859042`-`14859046` completed residual+Gaussian hybrids. K32/sigma0.35
92
+ reaches 31.30%; K64/sigma0.50 reaches 30.90%, so Gaussian hybridization is a
93
+ negative diagnostic.
94
+
95
+ Active residual-family mask jobs:
96
+
97
+ - Jobs `14859141`/`14859142` evaluate and summarize residual scale `0.50` while
98
+ excluding `residual_random_negative`.
99
+ - Jobs `14859143`/`14859144` evaluate and summarize residual scale `0.50` while
100
+ excluding `residual_random_negative` and `residual_wrong_direction`.
101
+ - Jobs `14859145`/`14859146` evaluate and summarize residual scale `0.25` with
102
+ the same random/wrong-direction mask.
103
+ - Jobs `14859147`/`14859148` evaluate and summarize residual scale `0.50` with
104
+ the policy/no-op/wrong-gripper residual families.
105
+ - Job `14859149` rebuilds `paper_table_status.*` after the masked summaries.
106
 
107
  Paper table automation:
108
 
results/dovla_cil_run_report_2026-06-27.md CHANGED
@@ -13,7 +13,7 @@ The deployed h=16 behavior-cloning policy remains weak, but the interventional f
13
  | Near-miss distillation policy, BC x5 | 28.29% | -1.45 pp | Stronger BC improves slightly but remains below baseline |
14
  | Near-miss proposal + field, best-policy checkpoint | 26.32% | -3.42 pp | Field scoring around the BC-selected checkpoint is worse |
15
  | Near-miss proposal + field, field checkpoint | 30.14% | +0.41 pp | Deployment-clean route begins to recover the mechanism |
16
- | Near-miss proposal + field, BC x5 field checkpoint | 32.93% | +3.19 pp | Best deployment-clean result so far (`k64_sigma0.50`) |
17
  | Lattice field selection, no expert candidates | 56.99% | +27.25 pp | Conservative, reviewer-safe result |
18
  | Lattice field selection, near-miss only | 55.94% | +26.20 pp | Minimal local counterfactual proposal family |
19
  | Lattice field selection, no expert and no near-miss | 25.57% | -4.17 pp | Confirms near-miss proposals carry the gain |
@@ -25,7 +25,15 @@ The deployed h=16 behavior-cloning policy remains weak, but the interventional f
25
  | Best non-expert proposal + field | 26.49% | -3.25 pp | Broad proposal-field bridge also fails |
26
  | Field-selected no-expert policy, seed-0 train map | 26.84% | -2.90 pp | Train-split field-teacher student fails |
27
  | Field-selected no-expert policy + field, seed-0 train map | 27.65% | -2.09 pp | Field scoring around that student remains below baseline |
28
- | Train-state residual retrieval | 32.12% | +2.38 pp | Positive clean bridge, below current clean best |
 
 
 
 
 
 
 
 
29
  | KNN train-state residual retrieval | 29.91% | +0.17 pp | More retrieved tangent neighborhoods dilute the signal |
30
 
31
  Canonical summaries:
@@ -46,8 +54,12 @@ Canonical summaries:
46
  - `results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_field_sweep_summary.md`
47
  - `results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_v2_summary.md`
48
  - `results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_knn4_v2_summary.md`
 
 
49
  - `results/h16_policy_ckpt_nonexpert_policy_bc5_summary.md`
50
  - `results/h16_policy_ckpt_nonexpert_policy_bc5_bestpt_field_sweep_summary.md`
 
 
51
  - `results/h16_field_optim_near_miss_policy_bc5_bestpt_s4_trust05_afterany_summary.md`
52
  - `results/paper_table_status.md`
53
  - `results/field_selected_target_map_summary.md`
@@ -57,7 +69,7 @@ Non-canonical/stale summaries:
57
  - `results/h16_lattice_no_near_miss_no_expert_summary.md` was produced before fixing Slurm comma handling for multi-type exclusion. Use the `_v2` file instead.
58
  - `results/h16_policy_ckpt_near_miss_policy_bc5_field_k32_sigma0p35_summary.md` is absent/incomplete because the last seed was canceled after the `best.pt` field-checkpoint variant superseded it.
59
 
60
- ## Current Update - 2026-06-28 05:03 UTC
61
 
62
  The old optimistic h=16 policy story is no longer the paper story. Direct BC
63
  rollout stayed near 30%, while the same-state no-expert lattice reached 56.99%.
@@ -79,21 +91,31 @@ Completed since the first report:
79
  - all split: 2,873 groups, 2,273 `near_miss`, 600 non-near-miss no-expert targets.
80
  - Train-split field-teacher distillation completed as a negative diagnostic:
81
  direct student 26.84%, field-guided best 27.65%.
82
-
83
- Active field-teacher jobs:
84
-
85
- - `14858450` completed all-split `field_selected_noexpert_bc5_allmap` training
86
- with 100% train/val target coverage for all three seeds.
87
- - `14858451`/`14858452` and `14858453`/`14858454` are pending priority to
88
- evaluate direct and field-guided allmap rollouts.
89
- - `14858455` rebuilds `paper_table_status.*` after the allmap summaries.
90
- - `retrieval_residual_scale` support was added after the 32.12% nearest
91
- residual result. Jobs `14858875`/`14858877`/`14858879`/`14858881` evaluate
92
- scales `0.25`, `0.50`, `0.75`, and `1.25`; summaries
93
- `14858876`/`14858878`/`14858880`/`14858882` and table rebuild `14858883`
94
- are queued behind them. CPU unit smoke `14858978` validates the scale path;
95
- earlier smoke attempts `14858889`/`14858894` caught and fixed two wiring bugs
96
- before rollout jobs started.
 
 
 
 
 
 
 
 
 
 
97
 
98
  Use `results/paper_table_status.md` as the canonical current table. The job
99
  table below is historical and includes statuses that were true at earlier
 
13
  | Near-miss distillation policy, BC x5 | 28.29% | -1.45 pp | Stronger BC improves slightly but remains below baseline |
14
  | Near-miss proposal + field, best-policy checkpoint | 26.32% | -3.42 pp | Field scoring around the BC-selected checkpoint is worse |
15
  | Near-miss proposal + field, field checkpoint | 30.14% | +0.41 pp | Deployment-clean route begins to recover the mechanism |
16
+ | Near-miss proposal + field, BC x5 field checkpoint | 32.93% | +3.19 pp | Strong deployment-clean bridge (`k64_sigma0.50`) |
17
  | Lattice field selection, no expert candidates | 56.99% | +27.25 pp | Conservative, reviewer-safe result |
18
  | Lattice field selection, near-miss only | 55.94% | +26.20 pp | Minimal local counterfactual proposal family |
19
  | Lattice field selection, no expert and no near-miss | 25.57% | -4.17 pp | Confirms near-miss proposals carry the gain |
 
25
  | Best non-expert proposal + field | 26.49% | -3.25 pp | Broad proposal-field bridge also fails |
26
  | Field-selected no-expert policy, seed-0 train map | 26.84% | -2.90 pp | Train-split field-teacher student fails |
27
  | Field-selected no-expert policy + field, seed-0 train map | 27.65% | -2.09 pp | Field scoring around that student remains below baseline |
28
+ | Field-selected no-expert policy, aligned allmap | 28.00% | -1.74 pp | 100% target-map coverage still fails |
29
+ | Field-selected no-expert policy + field, aligned allmap | 26.49% | -3.25 pp | Aligned field-teacher student remains below baseline |
30
+ | Train-state residual retrieval | 32.12% | +2.38 pp | Positive clean bridge |
31
+ | Train-state residual retrieval, scale 0.25 | 32.93% | +3.19 pp | Smaller tangent step ties prior clean best |
32
+ | Train-state residual retrieval, scale 0.50 | 33.33% | +3.59 pp | Current best deployment-clean bridge |
33
+ | Train-state residual retrieval, scale 0.75 | 32.70% | +2.96 pp | Larger tangent step is weaker |
34
+ | Train-state residual retrieval, scale 1.25 | 32.52% | +2.78 pp | Further scale increase does not help |
35
+ | Residual+Gaussian hybrid, K32 sigma0.35 | 31.30% | +1.57 pp | Hybrid proposals dilute residual transport |
36
+ | Residual+Gaussian hybrid, K64 sigma0.50 | 30.90% | +1.16 pp | Larger hybrid search is worse |
37
  | KNN train-state residual retrieval | 29.91% | +0.17 pp | More retrieved tangent neighborhoods dilute the signal |
38
 
39
  Canonical summaries:
 
54
  - `results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_field_sweep_summary.md`
55
  - `results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_v2_summary.md`
56
  - `results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_knn4_v2_summary.md`
57
+ - `results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_scale0p50_summary.md`
58
+ - `results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_hybrid_k32_sigma0p35_summary.md`
59
  - `results/h16_policy_ckpt_nonexpert_policy_bc5_summary.md`
60
  - `results/h16_policy_ckpt_nonexpert_policy_bc5_bestpt_field_sweep_summary.md`
61
+ - `results/h16_policy_ckpt_field_selected_noexpert_bc5_allmap_summary.md`
62
+ - `results/h16_policy_ckpt_field_selected_noexpert_bc5_allmap_bestpt_field_sweep_summary.md`
63
  - `results/h16_field_optim_near_miss_policy_bc5_bestpt_s4_trust05_afterany_summary.md`
64
  - `results/paper_table_status.md`
65
  - `results/field_selected_target_map_summary.md`
 
69
  - `results/h16_lattice_no_near_miss_no_expert_summary.md` was produced before fixing Slurm comma handling for multi-type exclusion. Use the `_v2` file instead.
70
  - `results/h16_policy_ckpt_near_miss_policy_bc5_field_k32_sigma0p35_summary.md` is absent/incomplete because the last seed was canceled after the `best.pt` field-checkpoint variant superseded it.
71
 
72
+ ## Current Update - 2026-06-28 05:47 UTC
73
 
74
  The old optimistic h=16 policy story is no longer the paper story. Direct BC
75
  rollout stayed near 30%, while the same-state no-expert lattice reached 56.99%.
 
91
  - all split: 2,873 groups, 2,273 `near_miss`, 600 non-near-miss no-expert targets.
92
  - Train-split field-teacher distillation completed as a negative diagnostic:
93
  direct student 26.84%, field-guided best 27.65%.
94
+ - Aligned all-split field-teacher distillation completed as a negative
95
+ diagnostic: direct student 28.00%, field-guided best 26.49%, despite 100%
96
+ target-map coverage.
97
+ - Residual scale sweep completed. Scale `0.50` is now the strongest
98
+ deployment-clean result at 33.33%; scale `0.25` ties the previous clean best
99
+ at 32.93%; scales `0.75` and `1.25` are weaker at 32.70% and 32.52%.
100
+ - Residual+Gaussian hybrids completed as negative diagnostics: K32/sigma0.35 is
101
+ 31.30%, K64/sigma0.50 is 30.90%.
102
+
103
+ Active residual-family mask jobs:
104
+
105
+ - `14859141`/`14859142`: scale `0.50`, excluding `residual_random_negative`.
106
+ - `14859143`/`14859144`: scale `0.50`, excluding `residual_random_negative` and
107
+ `residual_wrong_direction`.
108
+ - `14859145`/`14859146`: scale `0.25`, excluding the same anti-goal residuals.
109
+ - `14859147`/`14859148`: scale `0.50`, keeping policy/no-op/wrong-gripper
110
+ residual families.
111
+ - `14859149`: rebuilds `paper_table_status.*` after the masked summaries.
112
+
113
+ The motivation for this batch is the observed selected-candidate distribution
114
+ in the 33.33% scale-0.50 run: `residual_random_negative` and
115
+ `residual_wrong_direction` were selected 154 times across seeds and had low
116
+ conditional success. If masked residual transport improves, the clean story
117
+ becomes tighter: transferable local tangent proposals help only when the
118
+ proposal family is consistent with the counterfactual field's local geometry.
119
 
120
  Use `results/paper_table_status.md` as the canonical current table. The job
121
  table below is historical and includes statuses that were true at earlier
results/h16_policy_ckpt_field_selected_noexpert_bc5_allmap_bestpt_field_sweep_summary.json ADDED
@@ -0,0 +1,985 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "run_root": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_field_selected_noexpert_bc5_allmap_bestpt_field_sweep",
3
+ "baseline_h4_policy_success": 0.2967,
4
+ "baseline_h16_policy_success": 0.29739130434782607,
5
+ "num_result_files": 12,
6
+ "configs": [
7
+ {
8
+ "config": "k16_sigma0.20",
9
+ "completed_seeds": [
10
+ 0,
11
+ 1,
12
+ 2
13
+ ],
14
+ "num_completed": 3,
15
+ "mean_success": 0.2649275362318841,
16
+ "std_success": 0.02436852122548836,
17
+ "mean_progress": 0.49198602530857266,
18
+ "mean_action_mse_to_best": 0.48361168359986684,
19
+ "absolute_gain_vs_h4_baseline": -0.03177246376811593,
20
+ "absolute_gain_vs_h16_policy": -0.03246376811594198
21
+ },
22
+ {
23
+ "config": "k32_sigma0.35",
24
+ "completed_seeds": [
25
+ 0,
26
+ 1,
27
+ 2
28
+ ],
29
+ "num_completed": 3,
30
+ "mean_success": 0.263768115942029,
31
+ "std_success": 0.030931760166850078,
32
+ "mean_progress": 0.4907049547902231,
33
+ "mean_action_mse_to_best": 0.491781605578743,
34
+ "absolute_gain_vs_h4_baseline": -0.03293188405797104,
35
+ "absolute_gain_vs_h16_policy": -0.03362318840579709
36
+ },
37
+ {
38
+ "config": "k8_sigma0.10",
39
+ "completed_seeds": [
40
+ 0,
41
+ 1,
42
+ 2
43
+ ],
44
+ "num_completed": 3,
45
+ "mean_success": 0.263768115942029,
46
+ "std_success": 0.03208369646845615,
47
+ "mean_progress": 0.49222601328810267,
48
+ "mean_action_mse_to_best": 0.4790461642720291,
49
+ "absolute_gain_vs_h4_baseline": -0.03293188405797104,
50
+ "absolute_gain_vs_h16_policy": -0.03362318840579709
51
+ },
52
+ {
53
+ "config": "k64_sigma0.50",
54
+ "completed_seeds": [
55
+ 0,
56
+ 1,
57
+ 2
58
+ ],
59
+ "num_completed": 3,
60
+ "mean_success": 0.26318840579710145,
61
+ "std_success": 0.033422690563245386,
62
+ "mean_progress": 0.48753834678378444,
63
+ "mean_action_mse_to_best": 0.49955945385700984,
64
+ "absolute_gain_vs_h4_baseline": -0.033511594202898565,
65
+ "absolute_gain_vs_h16_policy": -0.034202898550724614
66
+ }
67
+ ],
68
+ "best": {
69
+ "config": "k16_sigma0.20",
70
+ "completed_seeds": [
71
+ 0,
72
+ 1,
73
+ 2
74
+ ],
75
+ "num_completed": 3,
76
+ "mean_success": 0.2649275362318841,
77
+ "std_success": 0.02436852122548836,
78
+ "mean_progress": 0.49198602530857266,
79
+ "mean_action_mse_to_best": 0.48361168359986684,
80
+ "absolute_gain_vs_h4_baseline": -0.03177246376811593,
81
+ "absolute_gain_vs_h16_policy": -0.03246376811594198
82
+ },
83
+ "rows": [
84
+ {
85
+ "config": "k16_sigma0.20",
86
+ "seed": 0,
87
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_field_selected_noexpert_bc5_allmap_bestpt_field_sweep/k16_sigma0.20/seed_0/online_rollout.json",
88
+ "selection_mode": "field",
89
+ "num_candidates": 16,
90
+ "candidate_sigma": 0.2,
91
+ "field_optim_steps": 0,
92
+ "field_optim_step_size": 0.0,
93
+ "field_optim_trust_radius": 0.0,
94
+ "field_optim_l2_penalty": 0.0,
95
+ "retrieval_neighbors": 0,
96
+ "num_groups": 575,
97
+ "policy_rollout_success_rate": 0.2573913043478261,
98
+ "policy_rollout_progress": 0.48223049327536266,
99
+ "oracle_success_rate": 0.8573913043478261,
100
+ "action_mse_to_best": 0.4800749153530468,
101
+ "per_task": {
102
+ "LiftPegUpright-v1": {
103
+ "action_mse_to_best": 0.337142610893652,
104
+ "expert_success_rate": 0.8865979381443299,
105
+ "num_groups": 97,
106
+ "oracle_success_rate": 0.9278350515463918,
107
+ "policy_expert_regret": 1.211708532627096,
108
+ "policy_oracle_regret": 1.2513229882286996,
109
+ "policy_rollout_progress": 0.49705433154229034,
110
+ "policy_rollout_success_rate": 0.15463917525773196,
111
+ "restore_max_error": 4.76837158203125e-07
112
+ },
113
+ "PickCube-v1": {
114
+ "action_mse_to_best": 0.44921097542660743,
115
+ "expert_success_rate": 0.9375,
116
+ "num_groups": 208,
117
+ "oracle_success_rate": 0.9471153846153846,
118
+ "policy_expert_regret": 1.2170655475266847,
119
+ "policy_oracle_regret": 1.2304884149531903,
120
+ "policy_rollout_progress": 0.49768345343986414,
121
+ "policy_rollout_success_rate": 0.1971153846153846,
122
+ "restore_max_error": 4.76837158203125e-07
123
+ },
124
+ "PullCube-v1": {
125
+ "action_mse_to_best": 0.7003706869028219,
126
+ "expert_success_rate": 0.19480519480519481,
127
+ "num_groups": 77,
128
+ "oracle_success_rate": 0.36363636363636365,
129
+ "policy_expert_regret": 0.23235792289306592,
130
+ "policy_oracle_regret": 0.47640881527747425,
131
+ "policy_rollout_progress": 0.2957020977197529,
132
+ "policy_rollout_success_rate": 0.14285714285714285,
133
+ "restore_max_error": 4.76837158203125e-07
134
+ },
135
+ "PushCube-v1": {
136
+ "action_mse_to_best": 0.41541208017696724,
137
+ "expert_success_rate": 0.8617021276595744,
138
+ "num_groups": 94,
139
+ "oracle_success_rate": 0.9893617021276596,
140
+ "policy_expert_regret": 0.5181357007711491,
141
+ "policy_oracle_regret": 0.6158189672104856,
142
+ "policy_rollout_progress": 0.6943406551442248,
143
+ "policy_rollout_success_rate": 0.6702127659574468,
144
+ "restore_max_error": 4.76837158203125e-07
145
+ },
146
+ "StackCube-v1": {
147
+ "action_mse_to_best": 0.5750210278225366,
148
+ "expert_success_rate": 0.6767676767676768,
149
+ "num_groups": 99,
150
+ "oracle_success_rate": 0.8585858585858586,
151
+ "policy_expert_regret": 1.0841046621402104,
152
+ "policy_oracle_regret": 1.272457669028128,
153
+ "policy_rollout_progress": 0.37891941464910606,
154
+ "policy_rollout_success_rate": 0.18181818181818182,
155
+ "restore_max_error": 3.948807716369629e-07
156
+ }
157
+ }
158
+ },
159
+ {
160
+ "config": "k16_sigma0.20",
161
+ "seed": 1,
162
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_field_selected_noexpert_bc5_allmap_bestpt_field_sweep/k16_sigma0.20/seed_1/online_rollout.json",
163
+ "selection_mode": "field",
164
+ "num_candidates": 16,
165
+ "candidate_sigma": 0.2,
166
+ "field_optim_steps": 0,
167
+ "field_optim_step_size": 0.0,
168
+ "field_optim_trust_radius": 0.0,
169
+ "field_optim_l2_penalty": 0.0,
170
+ "retrieval_neighbors": 0,
171
+ "num_groups": 575,
172
+ "policy_rollout_success_rate": 0.24521739130434783,
173
+ "policy_rollout_progress": 0.48871822979561885,
174
+ "oracle_success_rate": 0.8695652173913043,
175
+ "action_mse_to_best": 0.46061476999407874,
176
+ "per_task": {
177
+ "LiftPegUpright-v1": {
178
+ "action_mse_to_best": 0.34245977673547723,
179
+ "expert_success_rate": 0.8584070796460177,
180
+ "num_groups": 113,
181
+ "oracle_success_rate": 0.9380530973451328,
182
+ "policy_expert_regret": 1.1525215854686974,
183
+ "policy_oracle_regret": 1.2451238524069828,
184
+ "policy_rollout_progress": 0.5279117630118817,
185
+ "policy_rollout_success_rate": 0.1592920353982301,
186
+ "restore_max_error": 4.76837158203125e-07
187
+ },
188
+ "PickCube-v1": {
189
+ "action_mse_to_best": 0.3850769985710149,
190
+ "expert_success_rate": 0.9402173913043478,
191
+ "num_groups": 184,
192
+ "oracle_success_rate": 0.9456521739130435,
193
+ "policy_expert_regret": 1.262801991853532,
194
+ "policy_oracle_regret": 1.2703388968208815,
195
+ "policy_rollout_progress": 0.5141702485504104,
196
+ "policy_rollout_success_rate": 0.14673913043478262,
197
+ "restore_max_error": 4.76837158203125e-07
198
+ },
199
+ "PullCube-v1": {
200
+ "action_mse_to_best": 0.6737498645854526,
201
+ "expert_success_rate": 0.25,
202
+ "num_groups": 76,
203
+ "oracle_success_rate": 0.40789473684210525,
204
+ "policy_expert_regret": 0.39387192858971265,
205
+ "policy_oracle_regret": 0.6791693281725442,
206
+ "policy_rollout_progress": 0.22112822339036747,
207
+ "policy_rollout_success_rate": 0.07894736842105263,
208
+ "restore_max_error": 4.76837158203125e-07
209
+ },
210
+ "PushCube-v1": {
211
+ "action_mse_to_best": 0.4016580322572777,
212
+ "expert_success_rate": 0.8198198198198198,
213
+ "num_groups": 111,
214
+ "oracle_success_rate": 1.0,
215
+ "policy_expert_regret": 0.5245991288810163,
216
+ "policy_oracle_regret": 0.5908067684184324,
217
+ "policy_rollout_progress": 0.7154995378878739,
218
+ "policy_rollout_success_rate": 0.6936936936936937,
219
+ "restore_max_error": 4.76837158203125e-07
220
+ },
221
+ "StackCube-v1": {
222
+ "action_mse_to_best": 0.6539817466963451,
223
+ "expert_success_rate": 0.6923076923076923,
224
+ "num_groups": 91,
225
+ "oracle_success_rate": 0.8571428571428571,
226
+ "policy_expert_regret": 1.1689082289135062,
227
+ "policy_oracle_regret": 1.341029433431206,
228
+ "policy_rollout_progress": 0.33544432412792036,
229
+ "policy_rollout_success_rate": 0.14285714285714285,
230
+ "restore_max_error": 3.948807716369629e-07
231
+ }
232
+ }
233
+ },
234
+ {
235
+ "config": "k16_sigma0.20",
236
+ "seed": 2,
237
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_field_selected_noexpert_bc5_allmap_bestpt_field_sweep/k16_sigma0.20/seed_2/online_rollout.json",
238
+ "selection_mode": "field",
239
+ "num_candidates": 16,
240
+ "candidate_sigma": 0.2,
241
+ "field_optim_steps": 0,
242
+ "field_optim_step_size": 0.0,
243
+ "field_optim_trust_radius": 0.0,
244
+ "field_optim_l2_penalty": 0.0,
245
+ "retrieval_neighbors": 0,
246
+ "num_groups": 575,
247
+ "policy_rollout_success_rate": 0.2921739130434783,
248
+ "policy_rollout_progress": 0.5050093528547365,
249
+ "oracle_success_rate": 0.8765217391304347,
250
+ "action_mse_to_best": 0.5101453654524749,
251
+ "per_task": {
252
+ "LiftPegUpright-v1": {
253
+ "action_mse_to_best": 0.37076629759879626,
254
+ "expert_success_rate": 0.8229166666666666,
255
+ "num_groups": 96,
256
+ "oracle_success_rate": 0.9270833333333334,
257
+ "policy_expert_regret": 1.01052466314286,
258
+ "policy_oracle_regret": 1.1174735312039654,
259
+ "policy_rollout_progress": 0.5898309731855989,
260
+ "policy_rollout_success_rate": 0.23958333333333334,
261
+ "restore_max_error": 3.5762786865234375e-07
262
+ },
263
+ "PickCube-v1": {
264
+ "action_mse_to_best": 0.45198269856322293,
265
+ "expert_success_rate": 0.9444444444444444,
266
+ "num_groups": 198,
267
+ "oracle_success_rate": 0.9595959595959596,
268
+ "policy_expert_regret": 1.1855603451652434,
269
+ "policy_oracle_regret": 1.199689052740319,
270
+ "policy_rollout_progress": 0.5102074748667126,
271
+ "policy_rollout_success_rate": 0.23232323232323232,
272
+ "restore_max_error": 4.76837158203125e-07
273
+ },
274
+ "PullCube-v1": {
275
+ "action_mse_to_best": 0.7424894493901067,
276
+ "expert_success_rate": 0.24444444444444444,
277
+ "num_groups": 90,
278
+ "oracle_success_rate": 0.4666666666666667,
279
+ "policy_expert_regret": 0.36140653950472673,
280
+ "policy_oracle_regret": 0.6702879616369803,
281
+ "policy_rollout_progress": 0.2931647454284959,
282
+ "policy_rollout_success_rate": 0.13333333333333333,
283
+ "restore_max_error": 4.0978193283081055e-07
284
+ },
285
+ "PushCube-v1": {
286
+ "action_mse_to_best": 0.42433159990180835,
287
+ "expert_success_rate": 0.8514851485148515,
288
+ "num_groups": 101,
289
+ "oracle_success_rate": 1.0,
290
+ "policy_expert_regret": 0.4179915769560502,
291
+ "policy_oracle_regret": 0.5274729412971156,
292
+ "policy_rollout_progress": 0.7497547814751616,
293
+ "policy_rollout_success_rate": 0.7227722772277227,
294
+ "restore_max_error": 4.76837158203125e-07
295
+ },
296
+ "StackCube-v1": {
297
+ "action_mse_to_best": 0.6507322690553136,
298
+ "expert_success_rate": 0.7666666666666667,
299
+ "num_groups": 90,
300
+ "oracle_success_rate": 0.9111111111111111,
301
+ "policy_expert_regret": 1.2500304205550088,
302
+ "policy_oracle_regret": 1.3861764381329218,
303
+ "policy_rollout_progress": 0.34028293804989923,
304
+ "policy_rollout_success_rate": 0.15555555555555556,
305
+ "restore_max_error": 4.76837158203125e-07
306
+ }
307
+ }
308
+ },
309
+ {
310
+ "config": "k32_sigma0.35",
311
+ "seed": 0,
312
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_field_selected_noexpert_bc5_allmap_bestpt_field_sweep/k32_sigma0.35/seed_0/online_rollout.json",
313
+ "selection_mode": "field",
314
+ "num_candidates": 32,
315
+ "candidate_sigma": 0.35,
316
+ "field_optim_steps": 0,
317
+ "field_optim_step_size": 0.0,
318
+ "field_optim_trust_radius": 0.0,
319
+ "field_optim_l2_penalty": 0.0,
320
+ "retrieval_neighbors": 0,
321
+ "num_groups": 575,
322
+ "policy_rollout_success_rate": 0.2573913043478261,
323
+ "policy_rollout_progress": 0.4835339437833349,
324
+ "oracle_success_rate": 0.8573913043478261,
325
+ "action_mse_to_best": 0.4897524932399392,
326
+ "per_task": {
327
+ "LiftPegUpright-v1": {
328
+ "action_mse_to_best": 0.33845987016354334,
329
+ "expert_success_rate": 0.8865979381443299,
330
+ "num_groups": 97,
331
+ "oracle_success_rate": 0.9278350515463918,
332
+ "policy_expert_regret": 1.2120874862080997,
333
+ "policy_oracle_regret": 1.2517019418097033,
334
+ "policy_rollout_progress": 0.4966753779612866,
335
+ "policy_rollout_success_rate": 0.15463917525773196,
336
+ "restore_max_error": 4.76837158203125e-07
337
+ },
338
+ "PickCube-v1": {
339
+ "action_mse_to_best": 0.4614338276621241,
340
+ "expert_success_rate": 0.9375,
341
+ "num_groups": 208,
342
+ "oracle_success_rate": 0.9471153846153846,
343
+ "policy_expert_regret": 1.2167317788600065,
344
+ "policy_oracle_regret": 1.230154646286512,
345
+ "policy_rollout_progress": 0.49801722210654237,
346
+ "policy_rollout_success_rate": 0.1971153846153846,
347
+ "restore_max_error": 4.76837158203125e-07
348
+ },
349
+ "PullCube-v1": {
350
+ "action_mse_to_best": 0.7427825632855877,
351
+ "expert_success_rate": 0.19480519480519481,
352
+ "num_groups": 77,
353
+ "oracle_success_rate": 0.36363636363636365,
354
+ "policy_expert_regret": 0.2067434024065733,
355
+ "policy_oracle_regret": 0.4740380823805735,
356
+ "policy_rollout_progress": 0.30768172288095796,
357
+ "policy_rollout_success_rate": 0.15584415584415584,
358
+ "restore_max_error": 4.76837158203125e-07
359
+ },
360
+ "PushCube-v1": {
361
+ "action_mse_to_best": 0.41402394760479316,
362
+ "expert_success_rate": 0.8617021276595744,
363
+ "num_groups": 94,
364
+ "oracle_success_rate": 0.9893617021276596,
365
+ "policy_expert_regret": 0.4977186062234513,
366
+ "policy_oracle_regret": 0.6165330792044071,
367
+ "policy_rollout_progress": 0.6936265431503033,
368
+ "policy_rollout_success_rate": 0.6702127659574468,
369
+ "restore_max_error": 4.76837158203125e-07
370
+ },
371
+ "StackCube-v1": {
372
+ "action_mse_to_best": 0.5725892081367548,
373
+ "expert_success_rate": 0.6767676767676768,
374
+ "num_groups": 99,
375
+ "oracle_success_rate": 0.8585858585858586,
376
+ "policy_expert_regret": 1.0840482513109844,
377
+ "policy_oracle_regret": 1.272401258198902,
378
+ "policy_rollout_progress": 0.3775205666368658,
379
+ "policy_rollout_success_rate": 0.1717171717171717,
380
+ "restore_max_error": 3.948807716369629e-07
381
+ }
382
+ }
383
+ },
384
+ {
385
+ "config": "k32_sigma0.35",
386
+ "seed": 1,
387
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_field_selected_noexpert_bc5_allmap_bestpt_field_sweep/k32_sigma0.35/seed_1/online_rollout.json",
388
+ "selection_mode": "field",
389
+ "num_candidates": 32,
390
+ "candidate_sigma": 0.35,
391
+ "field_optim_steps": 0,
392
+ "field_optim_step_size": 0.0,
393
+ "field_optim_trust_radius": 0.0,
394
+ "field_optim_l2_penalty": 0.0,
395
+ "retrieval_neighbors": 0,
396
+ "num_groups": 575,
397
+ "policy_rollout_success_rate": 0.23652173913043478,
398
+ "policy_rollout_progress": 0.47912006016496733,
399
+ "oracle_success_rate": 0.8695652173913043,
400
+ "action_mse_to_best": 0.4697858330708645,
401
+ "per_task": {
402
+ "LiftPegUpright-v1": {
403
+ "action_mse_to_best": 0.3418705339040772,
404
+ "expert_success_rate": 0.8584070796460177,
405
+ "num_groups": 113,
406
+ "oracle_success_rate": 0.9380530973451328,
407
+ "policy_expert_regret": 1.1566909832500778,
408
+ "policy_oracle_regret": 1.2493307837079057,
409
+ "policy_rollout_progress": 0.5237048317109589,
410
+ "policy_rollout_success_rate": 0.1592920353982301,
411
+ "restore_max_error": 4.76837158203125e-07
412
+ },
413
+ "PickCube-v1": {
414
+ "action_mse_to_best": 0.3922307449595436,
415
+ "expert_success_rate": 0.9402173913043478,
416
+ "num_groups": 184,
417
+ "oracle_success_rate": 0.9456521739130435,
418
+ "policy_expert_regret": 1.2620471735920191,
419
+ "policy_oracle_regret": 1.2695840785593686,
420
+ "policy_rollout_progress": 0.5149250668119234,
421
+ "policy_rollout_success_rate": 0.14673913043478262,
422
+ "restore_max_error": 4.76837158203125e-07
423
+ },
424
+ "PullCube-v1": {
425
+ "action_mse_to_best": 0.7176830689821049,
426
+ "expert_success_rate": 0.25,
427
+ "num_groups": 76,
428
+ "oracle_success_rate": 0.40789473684210525,
429
+ "policy_expert_regret": 0.4004452299865845,
430
+ "policy_oracle_regret": 0.6885572214806942,
431
+ "policy_rollout_progress": 0.1979231177829206,
432
+ "policy_rollout_success_rate": 0.06578947368421052,
433
+ "restore_max_error": 4.76837158203125e-07
434
+ },
435
+ "PushCube-v1": {
436
+ "action_mse_to_best": 0.40793411527668033,
437
+ "expert_success_rate": 0.8198198198198198,
438
+ "num_groups": 111,
439
+ "oracle_success_rate": 1.0,
440
+ "policy_expert_regret": 0.5755485105621923,
441
+ "policy_oracle_regret": 0.6585513558742162,
442
+ "policy_rollout_progress": 0.6837909864681261,
443
+ "policy_rollout_success_rate": 0.6576576576576577,
444
+ "restore_max_error": 4.76837158203125e-07
445
+ },
446
+ "StackCube-v1": {
447
+ "action_mse_to_best": 0.6538508414689984,
448
+ "expert_success_rate": 0.6923076923076923,
449
+ "num_groups": 91,
450
+ "oracle_success_rate": 0.8571428571428571,
451
+ "policy_expert_regret": 1.1678006863528556,
452
+ "policy_oracle_regret": 1.3399218908705555,
453
+ "policy_rollout_progress": 0.3365518666885711,
454
+ "policy_rollout_success_rate": 0.14285714285714285,
455
+ "restore_max_error": 3.948807716369629e-07
456
+ }
457
+ }
458
+ },
459
+ {
460
+ "config": "k32_sigma0.35",
461
+ "seed": 2,
462
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_field_selected_noexpert_bc5_allmap_bestpt_field_sweep/k32_sigma0.35/seed_2/online_rollout.json",
463
+ "selection_mode": "field",
464
+ "num_candidates": 32,
465
+ "candidate_sigma": 0.35,
466
+ "field_optim_steps": 0,
467
+ "field_optim_step_size": 0.0,
468
+ "field_optim_trust_radius": 0.0,
469
+ "field_optim_l2_penalty": 0.0,
470
+ "retrieval_neighbors": 0,
471
+ "num_groups": 575,
472
+ "policy_rollout_success_rate": 0.29739130434782607,
473
+ "policy_rollout_progress": 0.5094608604223669,
474
+ "oracle_success_rate": 0.8765217391304347,
475
+ "action_mse_to_best": 0.5158064904254254,
476
+ "per_task": {
477
+ "LiftPegUpright-v1": {
478
+ "action_mse_to_best": 0.36748976713958353,
479
+ "expert_success_rate": 0.8229166666666666,
480
+ "num_groups": 96,
481
+ "oracle_success_rate": 0.9270833333333334,
482
+ "policy_expert_regret": 1.0264881861706574,
483
+ "policy_oracle_regret": 1.1334370542317629,
484
+ "policy_rollout_progress": 0.584284116824468,
485
+ "policy_rollout_success_rate": 0.22916666666666666,
486
+ "restore_max_error": 3.5762786865234375e-07
487
+ },
488
+ "PickCube-v1": {
489
+ "action_mse_to_best": 0.4534088330098776,
490
+ "expert_success_rate": 0.9444444444444444,
491
+ "num_groups": 198,
492
+ "oracle_success_rate": 0.9595959595959596,
493
+ "policy_expert_regret": 1.183910921390046,
494
+ "policy_oracle_regret": 1.1980396289651216,
495
+ "policy_rollout_progress": 0.5118568986419099,
496
+ "policy_rollout_success_rate": 0.23232323232323232,
497
+ "restore_max_error": 4.76837158203125e-07
498
+ },
499
+ "PullCube-v1": {
500
+ "action_mse_to_best": 0.777206671054268,
501
+ "expert_success_rate": 0.24444444444444444,
502
+ "num_groups": 90,
503
+ "oracle_success_rate": 0.4666666666666667,
504
+ "policy_expert_regret": 0.3903211276279762,
505
+ "policy_oracle_regret": 0.6822757954942062,
506
+ "policy_rollout_progress": 0.28765311417842493,
507
+ "policy_rollout_success_rate": 0.14444444444444443,
508
+ "restore_max_error": 4.0978193283081055e-07
509
+ },
510
+ "PushCube-v1": {
511
+ "action_mse_to_best": 0.4234293438125365,
512
+ "expert_success_rate": 0.8514851485148515,
513
+ "num_groups": 101,
514
+ "oracle_success_rate": 1.0,
515
+ "policy_expert_regret": 0.36308381831882025,
516
+ "policy_oracle_regret": 0.4722358460473542,
517
+ "policy_rollout_progress": 0.7752889064278933,
518
+ "policy_rollout_success_rate": 0.7524752475247525,
519
+ "restore_max_error": 4.76837158203125e-07
520
+ },
521
+ "StackCube-v1": {
522
+ "action_mse_to_best": 0.6535533477034834,
523
+ "expert_success_rate": 0.7666666666666667,
524
+ "num_groups": 90,
525
+ "oracle_success_rate": 0.9111111111111111,
526
+ "policy_expert_regret": 1.2424456500344807,
527
+ "policy_oracle_regret": 1.3785916676123937,
528
+ "policy_rollout_progress": 0.34786770857042737,
529
+ "policy_rollout_success_rate": 0.15555555555555556,
530
+ "restore_max_error": 4.76837158203125e-07
531
+ }
532
+ }
533
+ },
534
+ {
535
+ "config": "k64_sigma0.50",
536
+ "seed": 0,
537
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_field_selected_noexpert_bc5_allmap_bestpt_field_sweep/k64_sigma0.50/seed_0/online_rollout.json",
538
+ "selection_mode": "field",
539
+ "num_candidates": 64,
540
+ "candidate_sigma": 0.5,
541
+ "field_optim_steps": 0,
542
+ "field_optim_step_size": 0.0,
543
+ "field_optim_trust_radius": 0.0,
544
+ "field_optim_l2_penalty": 0.0,
545
+ "retrieval_neighbors": 0,
546
+ "num_groups": 575,
547
+ "policy_rollout_success_rate": 0.2573913043478261,
548
+ "policy_rollout_progress": 0.4811000149123696,
549
+ "oracle_success_rate": 0.8573913043478261,
550
+ "action_mse_to_best": 0.49755877656619185,
551
+ "per_task": {
552
+ "LiftPegUpright-v1": {
553
+ "action_mse_to_best": 0.3388602849499313,
554
+ "expert_success_rate": 0.8865979381443299,
555
+ "num_groups": 97,
556
+ "oracle_success_rate": 0.9278350515463918,
557
+ "policy_expert_regret": 1.2122231279451823,
558
+ "policy_oracle_regret": 1.2518375835467859,
559
+ "policy_rollout_progress": 0.496539736224204,
560
+ "policy_rollout_success_rate": 0.15463917525773196,
561
+ "restore_max_error": 4.76837158203125e-07
562
+ },
563
+ "PickCube-v1": {
564
+ "action_mse_to_best": 0.47225195370041406,
565
+ "expert_success_rate": 0.9375,
566
+ "num_groups": 208,
567
+ "oracle_success_rate": 0.9471153846153846,
568
+ "policy_expert_regret": 1.2168721025402187,
569
+ "policy_oracle_regret": 1.2302949699667243,
570
+ "policy_rollout_progress": 0.49787689842633015,
571
+ "policy_rollout_success_rate": 0.1971153846153846,
572
+ "restore_max_error": 4.76837158203125e-07
573
+ },
574
+ "PullCube-v1": {
575
+ "action_mse_to_best": 0.7871998975464662,
576
+ "expert_success_rate": 0.19480519480519481,
577
+ "num_groups": 77,
578
+ "oracle_success_rate": 0.36363636363636365,
579
+ "policy_expert_regret": 0.26621573498811235,
580
+ "policy_oracle_regret": 0.5400876995333314,
581
+ "policy_rollout_progress": 0.26673607360692575,
582
+ "policy_rollout_success_rate": 0.12987012987012986,
583
+ "restore_max_error": 4.76837158203125e-07
584
+ },
585
+ "PushCube-v1": {
586
+ "action_mse_to_best": 0.3990023007735293,
587
+ "expert_success_rate": 0.8617021276595744,
588
+ "num_groups": 94,
589
+ "oracle_success_rate": 0.9893617021276596,
590
+ "policy_expert_regret": 0.4784681630578447,
591
+ "policy_oracle_regret": 0.5775110150905366,
592
+ "policy_rollout_progress": 0.7113720115194929,
593
+ "policy_rollout_success_rate": 0.6914893617021277,
594
+ "restore_max_error": 4.76837158203125e-07
595
+ },
596
+ "StackCube-v1": {
597
+ "action_mse_to_best": 0.5745235770903151,
598
+ "expert_success_rate": 0.6767676767676768,
599
+ "num_groups": 99,
600
+ "oracle_success_rate": 0.8585858585858586,
601
+ "policy_expert_regret": 1.0827596003961082,
602
+ "policy_oracle_regret": 1.2711126072840258,
603
+ "policy_rollout_progress": 0.37880921755174196,
604
+ "policy_rollout_success_rate": 0.1717171717171717,
605
+ "restore_max_error": 3.948807716369629e-07
606
+ }
607
+ }
608
+ },
609
+ {
610
+ "config": "k64_sigma0.50",
611
+ "seed": 1,
612
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_field_selected_noexpert_bc5_allmap_bestpt_field_sweep/k64_sigma0.50/seed_1/online_rollout.json",
613
+ "selection_mode": "field",
614
+ "num_candidates": 64,
615
+ "candidate_sigma": 0.5,
616
+ "field_optim_steps": 0,
617
+ "field_optim_step_size": 0.0,
618
+ "field_optim_trust_radius": 0.0,
619
+ "field_optim_l2_penalty": 0.0,
620
+ "retrieval_neighbors": 0,
621
+ "num_groups": 575,
622
+ "policy_rollout_success_rate": 0.23304347826086957,
623
+ "policy_rollout_progress": 0.471685866382297,
624
+ "oracle_success_rate": 0.8695652173913043,
625
+ "action_mse_to_best": 0.4795833092899588,
626
+ "per_task": {
627
+ "LiftPegUpright-v1": {
628
+ "action_mse_to_best": 0.33992090817145276,
629
+ "expert_success_rate": 0.8584070796460177,
630
+ "num_groups": 113,
631
+ "oracle_success_rate": 0.9380530973451328,
632
+ "policy_expert_regret": 1.1292208964582038,
633
+ "policy_oracle_regret": 1.2218606969160317,
634
+ "policy_rollout_progress": 0.5334758034585851,
635
+ "policy_rollout_success_rate": 0.17699115044247787,
636
+ "restore_max_error": 4.76837158203125e-07
637
+ },
638
+ "PickCube-v1": {
639
+ "action_mse_to_best": 0.4006388466400297,
640
+ "expert_success_rate": 0.9402173913043478,
641
+ "num_groups": 184,
642
+ "oracle_success_rate": 0.9456521739130435,
643
+ "policy_expert_regret": 1.266307172653994,
644
+ "policy_oracle_regret": 1.2738440776213438,
645
+ "policy_rollout_progress": 0.5106650677499482,
646
+ "policy_rollout_success_rate": 0.14673913043478262,
647
+ "restore_max_error": 4.76837158203125e-07
648
+ },
649
+ "PullCube-v1": {
650
+ "action_mse_to_best": 0.7648348194259023,
651
+ "expert_success_rate": 0.25,
652
+ "num_groups": 76,
653
+ "oracle_success_rate": 0.40789473684210525,
654
+ "policy_expert_regret": 0.45474573209465136,
655
+ "policy_oracle_regret": 0.7474661598312914,
656
+ "policy_rollout_progress": 0.16442669933588294,
657
+ "policy_rollout_success_rate": 0.039473684210526314,
658
+ "restore_max_error": 4.76837158203125e-07
659
+ },
660
+ "PushCube-v1": {
661
+ "action_mse_to_best": 0.4144497260048583,
662
+ "expert_success_rate": 0.8198198198198198,
663
+ "num_groups": 111,
664
+ "oracle_success_rate": 1.0,
665
+ "policy_expert_regret": 0.6285274301294808,
666
+ "policy_oracle_regret": 0.6950307577042967,
667
+ "policy_rollout_progress": 0.6653296026560638,
668
+ "policy_rollout_success_rate": 0.6396396396396397,
669
+ "restore_max_error": 4.76837158203125e-07
670
+ },
671
+ "StackCube-v1": {
672
+ "action_mse_to_best": 0.6538508414689984,
673
+ "expert_success_rate": 0.6923076923076923,
674
+ "num_groups": 91,
675
+ "oracle_success_rate": 0.8571428571428571,
676
+ "policy_expert_regret": 1.1678006863528556,
677
+ "policy_oracle_regret": 1.3399218908705555,
678
+ "policy_rollout_progress": 0.3365518666885711,
679
+ "policy_rollout_success_rate": 0.14285714285714285,
680
+ "restore_max_error": 3.948807716369629e-07
681
+ }
682
+ }
683
+ },
684
+ {
685
+ "config": "k64_sigma0.50",
686
+ "seed": 2,
687
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_field_selected_noexpert_bc5_allmap_bestpt_field_sweep/k64_sigma0.50/seed_2/online_rollout.json",
688
+ "selection_mode": "field",
689
+ "num_candidates": 64,
690
+ "candidate_sigma": 0.5,
691
+ "field_optim_steps": 0,
692
+ "field_optim_step_size": 0.0,
693
+ "field_optim_trust_radius": 0.0,
694
+ "field_optim_l2_penalty": 0.0,
695
+ "retrieval_neighbors": 0,
696
+ "num_groups": 575,
697
+ "policy_rollout_success_rate": 0.2991304347826087,
698
+ "policy_rollout_progress": 0.5098291590566868,
699
+ "oracle_success_rate": 0.8765217391304347,
700
+ "action_mse_to_best": 0.5215362757148788,
701
+ "per_task": {
702
+ "LiftPegUpright-v1": {
703
+ "action_mse_to_best": 0.36730162321570486,
704
+ "expert_success_rate": 0.8229166666666666,
705
+ "num_groups": 96,
706
+ "oracle_success_rate": 0.9270833333333334,
707
+ "policy_expert_regret": 1.0263188638103504,
708
+ "policy_oracle_regret": 1.133267731871456,
709
+ "policy_rollout_progress": 0.584453439184775,
710
+ "policy_rollout_success_rate": 0.22916666666666666,
711
+ "restore_max_error": 3.5762786865234375e-07
712
+ },
713
+ "PickCube-v1": {
714
+ "action_mse_to_best": 0.4558377012121256,
715
+ "expert_success_rate": 0.9444444444444444,
716
+ "num_groups": 198,
717
+ "oracle_success_rate": 0.9595959595959596,
718
+ "policy_expert_regret": 1.183292665777521,
719
+ "policy_oracle_regret": 1.1903238530421272,
720
+ "policy_rollout_progress": 0.5145221695143993,
721
+ "policy_rollout_success_rate": 0.23737373737373738,
722
+ "restore_max_error": 4.76837158203125e-07
723
+ },
724
+ "PullCube-v1": {
725
+ "action_mse_to_best": 0.8013595571767331,
726
+ "expert_success_rate": 0.24444444444444444,
727
+ "num_groups": 90,
728
+ "oracle_success_rate": 0.4666666666666667,
729
+ "policy_expert_regret": 0.36839369783798853,
730
+ "policy_oracle_regret": 0.6607864346148239,
731
+ "policy_rollout_progress": 0.2989873856926958,
732
+ "policy_rollout_success_rate": 0.15555555555555556,
733
+ "restore_max_error": 4.0978193283081055e-07
734
+ },
735
+ "PushCube-v1": {
736
+ "action_mse_to_best": 0.4298998924470184,
737
+ "expert_success_rate": 0.8514851485148515,
738
+ "num_groups": 101,
739
+ "oracle_success_rate": 1.0,
740
+ "policy_expert_regret": 0.40163400798740956,
741
+ "policy_oracle_regret": 0.49443707888079164,
742
+ "policy_rollout_progress": 0.7629886636934659,
743
+ "policy_rollout_success_rate": 0.7425742574257426,
744
+ "restore_max_error": 4.76837158203125e-07
745
+ },
746
+ "StackCube-v1": {
747
+ "action_mse_to_best": 0.6536032064921327,
748
+ "expert_success_rate": 0.7666666666666667,
749
+ "num_groups": 90,
750
+ "oracle_success_rate": 0.9111111111111111,
751
+ "policy_expert_regret": 1.2436675031979878,
752
+ "policy_oracle_regret": 1.379813520775901,
753
+ "policy_rollout_progress": 0.3466458554069201,
754
+ "policy_rollout_success_rate": 0.15555555555555556,
755
+ "restore_max_error": 4.76837158203125e-07
756
+ }
757
+ }
758
+ },
759
+ {
760
+ "config": "k8_sigma0.10",
761
+ "seed": 0,
762
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_field_selected_noexpert_bc5_allmap_bestpt_field_sweep/k8_sigma0.10/seed_0/online_rollout.json",
763
+ "selection_mode": "field",
764
+ "num_candidates": 8,
765
+ "candidate_sigma": 0.1,
766
+ "field_optim_steps": 0,
767
+ "field_optim_step_size": 0.0,
768
+ "field_optim_trust_radius": 0.0,
769
+ "field_optim_l2_penalty": 0.0,
770
+ "retrieval_neighbors": 0,
771
+ "num_groups": 575,
772
+ "policy_rollout_success_rate": 0.25565217391304346,
773
+ "policy_rollout_progress": 0.4857378810579362,
774
+ "oracle_success_rate": 0.8573913043478261,
775
+ "action_mse_to_best": 0.47573888473212717,
776
+ "per_task": {
777
+ "LiftPegUpright-v1": {
778
+ "action_mse_to_best": 0.33964292481349606,
779
+ "expert_success_rate": 0.8865979381443299,
780
+ "num_groups": 97,
781
+ "oracle_success_rate": 0.9278350515463918,
782
+ "policy_expert_regret": 1.1928169364781724,
783
+ "policy_oracle_regret": 1.232431392079776,
784
+ "policy_rollout_progress": 0.5159459276912138,
785
+ "policy_rollout_success_rate": 0.15463917525773196,
786
+ "restore_max_error": 4.76837158203125e-07
787
+ },
788
+ "PickCube-v1": {
789
+ "action_mse_to_best": 0.4450738527095662,
790
+ "expert_success_rate": 0.9375,
791
+ "num_groups": 208,
792
+ "oracle_success_rate": 0.9471153846153846,
793
+ "policy_expert_regret": 1.2071357512434657,
794
+ "policy_oracle_regret": 1.2205586186699713,
795
+ "policy_rollout_progress": 0.5028055574153908,
796
+ "policy_rollout_success_rate": 0.20192307692307693,
797
+ "restore_max_error": 4.76837158203125e-07
798
+ },
799
+ "PullCube-v1": {
800
+ "action_mse_to_best": 0.691721789744477,
801
+ "expert_success_rate": 0.19480519480519481,
802
+ "num_groups": 77,
803
+ "oracle_success_rate": 0.36363636363636365,
804
+ "policy_expert_regret": 0.25618259988531666,
805
+ "policy_oracle_regret": 0.5514669314075213,
806
+ "policy_rollout_progress": 0.24744823137177274,
807
+ "policy_rollout_success_rate": 0.09090909090909091,
808
+ "restore_max_error": 4.76837158203125e-07
809
+ },
810
+ "PushCube-v1": {
811
+ "action_mse_to_best": 0.40548132902922784,
812
+ "expert_success_rate": 0.8617021276595744,
813
+ "num_groups": 94,
814
+ "oracle_success_rate": 0.9893617021276596,
815
+ "policy_expert_regret": 0.46030889213719267,
816
+ "policy_oracle_regret": 0.5593517874466613,
817
+ "policy_rollout_progress": 0.7188929412910279,
818
+ "policy_rollout_success_rate": 0.7021276595744681,
819
+ "restore_max_error": 4.76837158203125e-07
820
+ },
821
+ "StackCube-v1": {
822
+ "action_mse_to_best": 0.5722354637515364,
823
+ "expert_success_rate": 0.6767676767676768,
824
+ "num_groups": 99,
825
+ "oracle_success_rate": 0.8585858585858586,
826
+ "policy_expert_regret": 1.0773312383227878,
827
+ "policy_oracle_regret": 1.2656842452107053,
828
+ "policy_rollout_progress": 0.3842375796250623,
829
+ "policy_rollout_success_rate": 0.1717171717171717,
830
+ "restore_max_error": 3.948807716369629e-07
831
+ }
832
+ }
833
+ },
834
+ {
835
+ "config": "k8_sigma0.10",
836
+ "seed": 1,
837
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_field_selected_noexpert_bc5_allmap_bestpt_field_sweep/k8_sigma0.10/seed_1/online_rollout.json",
838
+ "selection_mode": "field",
839
+ "num_candidates": 8,
840
+ "candidate_sigma": 0.1,
841
+ "field_optim_steps": 0,
842
+ "field_optim_step_size": 0.0,
843
+ "field_optim_trust_radius": 0.0,
844
+ "field_optim_l2_penalty": 0.0,
845
+ "retrieval_neighbors": 0,
846
+ "num_groups": 575,
847
+ "policy_rollout_success_rate": 0.23652173913043478,
848
+ "policy_rollout_progress": 0.4856837678530618,
849
+ "oracle_success_rate": 0.8695652173913043,
850
+ "action_mse_to_best": 0.45626937851998145,
851
+ "per_task": {
852
+ "LiftPegUpright-v1": {
853
+ "action_mse_to_best": 0.34486329094795265,
854
+ "expert_success_rate": 0.8584070796460177,
855
+ "num_groups": 113,
856
+ "oracle_success_rate": 0.9380530973451328,
857
+ "policy_expert_regret": 1.1555887038728831,
858
+ "policy_oracle_regret": 1.2368099750143238,
859
+ "policy_rollout_progress": 0.5386279850406984,
860
+ "policy_rollout_success_rate": 0.168141592920354,
861
+ "restore_max_error": 4.76837158203125e-07
862
+ },
863
+ "PickCube-v1": {
864
+ "action_mse_to_best": 0.3821941536810735,
865
+ "expert_success_rate": 0.9402173913043478,
866
+ "num_groups": 184,
867
+ "oracle_success_rate": 0.9456521739130435,
868
+ "policy_expert_regret": 1.2916925651851394,
869
+ "policy_oracle_regret": 1.299229470152489,
870
+ "policy_rollout_progress": 0.5070188056535857,
871
+ "policy_rollout_success_rate": 0.125,
872
+ "restore_max_error": 4.76837158203125e-07
873
+ },
874
+ "PullCube-v1": {
875
+ "action_mse_to_best": 0.650797848899138,
876
+ "expert_success_rate": 0.25,
877
+ "num_groups": 76,
878
+ "oracle_success_rate": 0.40789473684210525,
879
+ "policy_expert_regret": 0.4193746543452634,
880
+ "policy_oracle_regret": 0.6760359260216845,
881
+ "policy_rollout_progress": 0.21423209038881683,
882
+ "policy_rollout_success_rate": 0.06578947368421052,
883
+ "restore_max_error": 4.76837158203125e-07
884
+ },
885
+ "PushCube-v1": {
886
+ "action_mse_to_best": 0.3952832786819419,
887
+ "expert_success_rate": 0.8198198198198198,
888
+ "num_groups": 111,
889
+ "oracle_success_rate": 1.0,
890
+ "policy_expert_regret": 0.5248040548852972,
891
+ "policy_oracle_regret": 0.626877814262837,
892
+ "policy_rollout_progress": 0.6974465100614874,
893
+ "policy_rollout_success_rate": 0.6756756756756757,
894
+ "restore_max_error": 4.76837158203125e-07
895
+ },
896
+ "StackCube-v1": {
897
+ "action_mse_to_best": 0.6563135829068475,
898
+ "expert_success_rate": 0.6923076923076923,
899
+ "num_groups": 91,
900
+ "oracle_success_rate": 0.8571428571428571,
901
+ "policy_expert_regret": 1.148159778543881,
902
+ "policy_oracle_regret": 1.320280983061581,
903
+ "policy_rollout_progress": 0.3452037635085347,
904
+ "policy_rollout_success_rate": 0.15384615384615385,
905
+ "restore_max_error": 3.948807716369629e-07
906
+ }
907
+ }
908
+ },
909
+ {
910
+ "config": "k8_sigma0.10",
911
+ "seed": 2,
912
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_field_selected_noexpert_bc5_allmap_bestpt_field_sweep/k8_sigma0.10/seed_2/online_rollout.json",
913
+ "selection_mode": "field",
914
+ "num_candidates": 8,
915
+ "candidate_sigma": 0.1,
916
+ "field_optim_steps": 0,
917
+ "field_optim_step_size": 0.0,
918
+ "field_optim_trust_radius": 0.0,
919
+ "field_optim_l2_penalty": 0.0,
920
+ "retrieval_neighbors": 0,
921
+ "num_groups": 575,
922
+ "policy_rollout_success_rate": 0.2991304347826087,
923
+ "policy_rollout_progress": 0.50525639095331,
924
+ "oracle_success_rate": 0.8765217391304347,
925
+ "action_mse_to_best": 0.5051302295639787,
926
+ "per_task": {
927
+ "LiftPegUpright-v1": {
928
+ "action_mse_to_best": 0.37213402109046,
929
+ "expert_success_rate": 0.8229166666666666,
930
+ "num_groups": 96,
931
+ "oracle_success_rate": 0.9270833333333334,
932
+ "policy_expert_regret": 1.068213837997367,
933
+ "policy_oracle_regret": 1.1885378474059205,
934
+ "policy_rollout_progress": 0.5471540091869732,
935
+ "policy_rollout_success_rate": 0.19791666666666666,
936
+ "restore_max_error": 3.5762786865234375e-07
937
+ },
938
+ "PickCube-v1": {
939
+ "action_mse_to_best": 0.45110814376837677,
940
+ "expert_success_rate": 0.9444444444444444,
941
+ "num_groups": 198,
942
+ "oracle_success_rate": 0.9595959595959596,
943
+ "policy_expert_regret": 1.17821413813855,
944
+ "policy_oracle_regret": 1.1923428457136256,
945
+ "policy_rollout_progress": 0.5125031768429008,
946
+ "policy_rollout_success_rate": 0.23737373737373738,
947
+ "restore_max_error": 4.76837158203125e-07
948
+ },
949
+ "PullCube-v1": {
950
+ "action_mse_to_best": 0.7196254622915552,
951
+ "expert_success_rate": 0.24444444444444444,
952
+ "num_groups": 90,
953
+ "oracle_success_rate": 0.4666666666666667,
954
+ "policy_expert_regret": 0.3943389943731341,
955
+ "policy_oracle_regret": 0.6857609202705337,
956
+ "policy_rollout_progress": 0.28567321347767977,
957
+ "policy_rollout_success_rate": 0.14444444444444443,
958
+ "restore_max_error": 4.0978193283081055e-07
959
+ },
960
+ "PushCube-v1": {
961
+ "action_mse_to_best": 0.4191755559479836,
962
+ "expert_success_rate": 0.8514851485148515,
963
+ "num_groups": 101,
964
+ "oracle_success_rate": 1.0,
965
+ "policy_expert_regret": 0.37985018205524673,
966
+ "policy_oracle_regret": 0.43452872099852796,
967
+ "policy_rollout_progress": 0.7931940512786998,
968
+ "policy_rollout_success_rate": 0.7722772277227723,
969
+ "restore_max_error": 4.76837158203125e-07
970
+ },
971
+ "StackCube-v1": {
972
+ "action_mse_to_best": 0.6478064527942075,
973
+ "expert_success_rate": 0.7666666666666667,
974
+ "num_groups": 90,
975
+ "oracle_success_rate": 0.9111111111111111,
976
+ "policy_expert_regret": 1.2381264418363571,
977
+ "policy_oracle_regret": 1.3742724594142701,
978
+ "policy_rollout_progress": 0.3410758056574398,
979
+ "policy_rollout_success_rate": 0.16666666666666666,
980
+ "restore_max_error": 4.76837158203125e-07
981
+ }
982
+ }
983
+ }
984
+ ]
985
+ }
results/h16_policy_ckpt_field_selected_noexpert_bc5_allmap_bestpt_field_sweep_summary.md ADDED
@@ -0,0 +1,18 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # h=16 Field-Guided Rollout Sweep
2
+
3
+ Result root: `/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_field_selected_noexpert_bc5_allmap_bestpt_field_sweep`
4
+ Completed result files: 12
5
+ Baseline h=4 policy success: 29.67%
6
+ Baseline h=16 policy success: 29.74%
7
+
8
+ | config | seeds | mean success | gain vs h=16 | progress | action MSE |
9
+ |---|---:|---:|---:|---:|---:|
10
+ | k16_sigma0.20 | 3 | 26.49% | -3.25% | 49.20% | 0.484 |
11
+ | k32_sigma0.35 | 3 | 26.38% | -3.36% | 49.07% | 0.492 |
12
+ | k8_sigma0.10 | 3 | 26.38% | -3.36% | 49.22% | 0.479 |
13
+ | k64_sigma0.50 | 3 | 26.32% | -3.42% | 48.75% | 0.500 |
14
+
15
+ Best config:
16
+ - k16_sigma0.20
17
+ - mean success: 26.49%
18
+ - gain vs h=16 policy: -3.25%