Auto-sync: 2026-06-28 01:40:05

results/README.md

@@ -24,6 +24,8 @@ Use these summaries for the current paper tables:
 - `h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_knn4_v2_summary.md`
 - pending scale-sweep summaries:
   `h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_scale{0p25,0p50,0p75,1p25}_summary.md`
+- pending hybrid summaries:
+  `h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_hybrid_k{32,64}_sigma*.md`
 - `h16_policy_ckpt_nonexpert_policy_bc5_summary.md`
 - `h16_policy_ckpt_nonexpert_policy_bc5_bestpt_field_sweep_summary.md`
 - `h16_policy_ckpt_field_selected_noexpert_bc5_summary.md`
@@ -85,6 +87,11 @@ Active residual transport jobs:
   `14858879`/`14858880`, and `14858881`/`14858882` evaluate residual scales
   `0.25`, `0.50`, `0.75`, and `1.25`.
 - Job `14858883` rebuilds `paper_table_status.*` after the scale summaries.
+- Job `14859041` completed the CPU Apptainer unit smoke for hybrid
+  residual+Gaussian selection.
+- Jobs `14859042`/`14859043` and `14859044`/`14859045` evaluate residual+Gaussian
+  hybrid proposals at K32/sigma0.35 and K64/sigma0.50.
+- Job `14859046` rebuilds `paper_table_status.*` after the hybrid summaries.
 
 Paper table automation:
 

results/h16_policy_ckpt_field_selected_noexpert_bc5_allmap_summary.json

@@ -0,0 +1,277 @@
+{
+  "run_root": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs",
+  "objective": "field_selected_noexpert_bc5_allmap",
+  "out_name": "policy_rollout.json",
+  "num_completed": 3,
+  "baseline_h4_policy_success": 0.2967,
+  "baseline_h16_rank_checkpoint_success": 0.29739130434782607,
+  "mean_success": 0.27999999999999997,
+  "std_success": 0.06683783948026308,
+  "mean_progress": 0.5102699262808628,
+  "mean_action_mse_to_best": 0.46968574777866406,
+  "gain_vs_h4": -0.016700000000000048,
+  "gain_vs_h16_rank_checkpoint": -0.017391304347826098,
+  "rows": [
+    {
+      "seed": 0,
+      "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/field_selected_noexpert_bc5_allmap/seed_0/policy_rollout.json",
+      "num_groups": 575,
+      "selection_mode": "policy",
+      "num_candidates": 1,
+      "candidate_sigma": 0.0,
+      "field_optim_steps": 0,
+      "field_optim_step_size": 0.0,
+      "field_optim_trust_radius": 0.0,
+      "field_optim_l2_penalty": 0.0,
+      "retrieval_neighbors": 0,
+      "policy_rollout_success_rate": 0.20347826086956522,
+      "policy_rollout_progress": 0.46757474173670227,
+      "oracle_success_rate": 0.8573913043478261,
+      "action_mse_to_best": 0.4444025756256736,
+      "best_policy_val": {
+        "bc_loss": 0.17165934460030663,
+        "field_effect_loss": 0.010001767168028487,
+        "field_potential_loss": 0.26052173723777133,
+        "field_preference_loss": 0.5052156729830636,
+        "lattice_edges": 3833.3333333333335,
+        "progress_mae": 0.1823364802532726,
+        "rank_acc": 0.830044021209081,
+        "rank_loss": 0.5052017254961861,
+        "regret_mae": 0.4112023413181305,
+        "success_accuracy": 0.8822524613804288,
+        "total_loss": 1.0269725090927548
+      },
+      "per_task": {
+        "LiftPegUpright-v1": {
+          "action_mse_to_best": 0.3459187798470873,
+          "expert_success_rate": 0.8865979381443299,
+          "num_groups": 97,
+          "oracle_success_rate": 0.9278350515463918,
+          "policy_expert_regret": 1.296359863477884,
+          "policy_oracle_regret": 1.3359743190794875,
+          "policy_rollout_progress": 0.46680313318046096,
+          "policy_rollout_success_rate": 0.1134020618556701,
+          "restore_max_error": 4.76837158203125e-07
+        },
+        "PickCube-v1": {
+          "action_mse_to_best": 0.36527677620044696,
+          "expert_success_rate": 0.9375,
+          "num_groups": 208,
+          "oracle_success_rate": 0.9471153846153846,
+          "policy_expert_regret": 1.3078391076459621,
+          "policy_oracle_regret": 1.3145056817000016,
+          "policy_rollout_progress": 0.4877222910917436,
+          "policy_rollout_success_rate": 0.12980769230769232,
+          "restore_max_error": 4.76837158203125e-07
+        },
+        "PullCube-v1": {
+          "action_mse_to_best": 0.6650680970981137,
+          "expert_success_rate": 0.19480519480519481,
+          "num_groups": 77,
+          "oracle_success_rate": 0.36363636363636365,
+          "policy_expert_regret": 0.3585849960225743,
+          "policy_oracle_regret": 0.6478847708891738,
+          "policy_rollout_progress": 0.21878486915261713,
+          "policy_rollout_success_rate": 0.05194805194805195,
+          "restore_max_error": 4.76837158203125e-07
+        },
+        "PushCube-v1": {
+          "action_mse_to_best": 0.3953499881986608,
+          "expert_success_rate": 0.8617021276595744,
+          "num_groups": 94,
+          "oracle_success_rate": 0.9893617021276596,
+          "policy_expert_regret": 0.5183219870037221,
+          "policy_oracle_regret": 0.6172864880967648,
+          "policy_rollout_progress": 0.693265235170405,
+          "policy_rollout_success_rate": 0.6702127659574468,
+          "restore_max_error": 4.76837158203125e-07
+        },
+        "StackCube-v1": {
+          "action_mse_to_best": 0.5820873487138688,
+          "expert_success_rate": 0.6767676767676768,
+          "num_groups": 99,
+          "oracle_success_rate": 0.8585858585858586,
+          "policy_expert_regret": 1.095356533021638,
+          "policy_oracle_regret": 1.295081411949312,
+          "policy_rollout_progress": 0.4052118081035036,
+          "policy_rollout_success_rate": 0.12121212121212122,
+          "restore_max_error": 3.948807716369629e-07
+        }
+      }
+    },
+    {
+      "seed": 1,
+      "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/field_selected_noexpert_bc5_allmap/seed_1/policy_rollout.json",
+      "num_groups": 575,
+      "selection_mode": "policy",
+      "num_candidates": 1,
+      "candidate_sigma": 0.0,
+      "field_optim_steps": 0,
+      "field_optim_step_size": 0.0,
+      "field_optim_trust_radius": 0.0,
+      "field_optim_l2_penalty": 0.0,
+      "retrieval_neighbors": 0,
+      "policy_rollout_success_rate": 0.3095652173913043,
+      "policy_rollout_progress": 0.5244166037255649,
+      "oracle_success_rate": 0.8695652173913043,
+      "action_mse_to_best": 0.4580515311949927,
+      "best_policy_val": {
+        "bc_loss": 0.16598059071434867,
+        "field_effect_loss": 0.009884200886719756,
+        "field_potential_loss": 0.2825224987334675,
+        "field_preference_loss": 0.5118036468823751,
+        "lattice_edges": 3833.3333333333335,
+        "progress_mae": 0.19128910534911686,
+        "rank_acc": 0.8297591176297929,
+        "rank_loss": 0.5117973801162508,
+        "regret_mae": 0.36752037041717106,
+        "success_accuracy": 0.8753675222396851,
+        "total_loss": 1.0538421769936879
+      },
+      "per_task": {
+        "LiftPegUpright-v1": {
+          "action_mse_to_best": 0.3440804638388872,
+          "expert_success_rate": 0.8584070796460177,
+          "num_groups": 113,
+          "oracle_success_rate": 0.9380530973451328,
+          "policy_expert_regret": 1.0801474873998524,
+          "policy_oracle_regret": 1.171662218802798,
+          "policy_rollout_progress": 0.5595992191702919,
+          "policy_rollout_success_rate": 0.21238938053097345,
+          "restore_max_error": 4.76837158203125e-07
+        },
+        "PickCube-v1": {
+          "action_mse_to_best": 0.40015124578190886,
+          "expert_success_rate": 0.9402173913043478,
+          "num_groups": 184,
+          "oracle_success_rate": 0.9456521739130435,
+          "policy_expert_regret": 1.1359283014666308,
+          "policy_oracle_regret": 1.1434652064339803,
+          "policy_rollout_progress": 0.5430445673490815,
+          "policy_rollout_success_rate": 0.24456521739130435,
+          "restore_max_error": 4.76837158203125e-07
+        },
+        "PullCube-v1": {
+          "action_mse_to_best": 0.663783856453129,
+          "expert_success_rate": 0.25,
+          "num_groups": 76,
+          "oracle_success_rate": 0.40789473684210525,
+          "policy_expert_regret": 0.42638231974438223,
+          "policy_oracle_regret": 0.7047497212862301,
+          "policy_rollout_progress": 0.19452233491220364,
+          "policy_rollout_success_rate": 0.05263157894736842,
+          "restore_max_error": 4.76837158203125e-07
+        },
+        "PushCube-v1": {
+          "action_mse_to_best": 0.3955722923415738,
+          "expert_success_rate": 0.8198198198198198,
+          "num_groups": 111,
+          "oracle_success_rate": 1.0,
+          "policy_expert_regret": 0.29465652116247126,
+          "policy_oracle_regret": 0.3455311407645543,
+          "policy_rollout_progress": 0.8346490394156258,
+          "policy_rollout_success_rate": 0.8198198198198198,
+          "restore_max_error": 4.76837158203125e-07
+        },
+        "StackCube-v1": {
+          "action_mse_to_best": 0.6210396841659651,
+          "expert_success_rate": 0.6923076923076923,
+          "num_groups": 91,
+          "oracle_success_rate": 0.8571428571428571,
+          "policy_expert_regret": 1.1406350975835717,
+          "policy_oracle_regret": 1.3253210745342485,
+          "policy_rollout_progress": 0.34016367203586706,
+          "policy_rollout_success_rate": 0.15384615384615385,
+          "restore_max_error": 3.948807716369629e-07
+        }
+      }
+    },
+    {
+      "seed": 2,
+      "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/field_selected_noexpert_bc5_allmap/seed_2/policy_rollout.json",
+      "num_groups": 575,
+      "selection_mode": "policy",
+      "num_candidates": 1,
+      "candidate_sigma": 0.0,
+      "field_optim_steps": 0,
+      "field_optim_step_size": 0.0,
+      "field_optim_trust_radius": 0.0,
+      "field_optim_l2_penalty": 0.0,
+      "retrieval_neighbors": 0,
+      "policy_rollout_success_rate": 0.3269565217391304,
+      "policy_rollout_progress": 0.5388184333803213,
+      "oracle_success_rate": 0.8765217391304347,
+      "action_mse_to_best": 0.5066031365153258,
+      "best_policy_val": {
+        "bc_loss": 0.15605498850345612,
+        "field_effect_loss": 0.009339532194038233,
+        "field_potential_loss": 0.2986210021707747,
+        "field_preference_loss": 0.5134152736928728,
+        "lattice_edges": 3833.3333333333335,
+        "progress_mae": 0.19033906939956877,
+        "rank_acc": 0.8259397314654456,
+        "rank_loss": 0.5133909400966432,
+        "regret_mae": 0.4042019529475106,
+        "success_accuracy": 0.8761165704992082,
+        "total_loss": 1.0592188636461894
+      },
+      "per_task": {
+        "LiftPegUpright-v1": {
+          "action_mse_to_best": 0.36337641860397224,
+          "expert_success_rate": 0.8229166666666666,
+          "num_groups": 96,
+          "oracle_success_rate": 0.9270833333333334,
+          "policy_expert_regret": 1.0439447644166648,
+          "policy_oracle_regret": 1.1644653449766338,
+          "policy_rollout_progress": 0.5683207125402987,
+          "policy_rollout_success_rate": 0.1875,
+          "restore_max_error": 3.5762786865234375e-07
+        },
+        "PickCube-v1": {
+          "action_mse_to_best": 0.4642809172786509,
+          "expert_success_rate": 0.9444444444444444,
+          "num_groups": 198,
+          "oracle_success_rate": 0.9595959595959596,
+          "policy_expert_regret": 0.9911650435650966,
+          "policy_oracle_regret": 0.9982484940938049,
+          "policy_rollout_progress": 0.6059832450406946,
+          "policy_rollout_success_rate": 0.3434343434343434,
+          "restore_max_error": 4.76837158203125e-07
+        },
+        "PullCube-v1": {
+          "action_mse_to_best": 0.7153139960331222,
+          "expert_success_rate": 0.24444444444444444,
+          "num_groups": 90,
+          "oracle_success_rate": 0.4666666666666667,
+          "policy_expert_regret": 0.38542827831374277,
+          "policy_oracle_regret": 0.7176565832561916,
+          "policy_rollout_progress": 0.24548330381512642,
+          "policy_rollout_success_rate": 0.1111111111111111,
+          "restore_max_error": 4.0978193283081055e-07
+        },
+        "PushCube-v1": {
+          "action_mse_to_best": 0.42691447681719713,
+          "expert_success_rate": 0.8514851485148515,
+          "num_groups": 101,
+          "oracle_success_rate": 1.0,
+          "policy_expert_regret": 0.337999097043925,
+          "policy_oracle_regret": 0.3737342174690549,
+          "policy_rollout_progress": 0.8242855845111432,
+          "policy_rollout_success_rate": 0.801980198019802,
+          "restore_max_error": 4.76837158203125e-07
+        },
+        "StackCube-v1": {
+          "action_mse_to_best": 0.6332047098626693,
+          "expert_success_rate": 0.7666666666666667,
+          "num_groups": 90,
+          "oracle_success_rate": 0.9111111111111111,
+          "policy_expert_regret": 1.278331852455934,
+          "policy_oracle_regret": 1.4272284107075797,
+          "policy_rollout_progress": 0.3325642988085747,
+          "policy_rollout_success_rate": 0.12222222222222222,
+          "restore_max_error": 4.76837158203125e-07
+        }
+      }
+    }
+  ]
+}

results/h16_policy_ckpt_field_selected_noexpert_bc5_allmap_summary.md

@@ -0,0 +1,19 @@
+# h=16 Best-Policy Checkpoint Rollout
+
+Run root: `/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs`
+Objective: `field_selected_noexpert_bc5_allmap`
+Result file: `policy_rollout.json`
+Completed seeds: 3
+Baseline h=4 policy success: 29.67%
+Baseline h=16 rank-checkpoint success: 29.74%
+
+Mean success: 28.00% +/- 6.68%
+Gain vs h=16 rank checkpoint: -1.74%
+Mean progress: 51.03%
+Mean action MSE to best: 0.470
+
+| seed | mode | k | retrieval K | sigma | opt steps | trust | success | progress | oracle | action MSE |
+|---:|---|---:|---:|---:|---:|---:|---:|---:|---:|---:|
+| 0 | policy | 1 | 0 | 0.00 | 0 | 0.00 | 20.35% | 46.76% | 85.74% | 0.444 |
+| 1 | policy | 1 | 0 | 0.00 | 0 | 0.00 | 30.96% | 52.44% | 86.96% | 0.458 |
+| 2 | policy | 1 | 0 | 0.00 | 0 | 0.00 | 32.70% | 53.88% | 87.65% | 0.507 |