cbb
219ed19de4
轮次51 (2x2 缺失格 K16+pre)：最好格 到达 0.5664 / 时限内 0.3594 / 2s误差比 0.1382 / 行程比 2.4142 / 碰撞 0 / 超VMAX 0。主效应：pre 在 K8 +6.64pp、K16 +6.25pp(时限)， K16 在 endpoint +3.13pp、pre +2.73pp ⇒ 时限可加无交互；到达饱和(pre 在 K16 上净+1题 p=1.0)。 最好格 vs 历史控制逐题赢20丢9 净+11，McNemar p=0.061 ⇒ 到达仍未确立。 轮次52 (仪表)：dump 补 t_exp/tr_ref(题目属性，与臂无关 ⇒ 一条臂即可给全部历史臂补配对)。 改完落盘代码先自证行为未动：同配方控制臂 22/22 老键逐位相同；limit=2*t_exp 实测对上 ⇒ "到达率自带 2x 时限"从推断变成实测。 逐题配对(vs K8 endpoint, 判据=ok 且 t_done<=1.3*t_exp)： K16pre 27赢3丢 净+24 p=8.4e-6；K8pre 22/5 净+17 p=0.0015；门控5 23/6 p=0.0023； 门控20 22/6 p=0.0037；h=4 15/5 p=0.041；K16endpoint 17/9 p=0.169；h=8 11/7 p=0.481； tie=1 3/0 p=0.25。 ⇒ 自我纠正：轮次46-51 我一直拿 success(那道 2x 门)做配对，把"时限内"写成没有配对支持。 现在两栏分开：到达未确立(p>=0.06)，时限内已确立(p=0.0015/8.4e-6)。 副产物：K16+endpoint 的 +3.1pp 时限当初没有配对检验，实测 p=0.169 不显著，已在轮次52记下。
111 lines
2.6 KiB
JSON
111 lines
2.6 KiB
JSON
{
|
||
"config": {
|
||
"ckpt": "ml/runs/base_dag2_dag1/best.pt",
|
||
"samples": "ml/data/samples",
|
||
"tasks": null,
|
||
"split": "val",
|
||
"n": 256,
|
||
"K": 8,
|
||
"commit": 2,
|
||
"fb": 2.2,
|
||
"ki": 0.0,
|
||
"fb_scale": 0.0,
|
||
"fb_near": 0.2,
|
||
"fb_fallback": 0.0,
|
||
"kdecay": 0.9,
|
||
"kimax": 0.5,
|
||
"fb_every": 1,
|
||
"safety_try": 0,
|
||
"sel_w": 0.1,
|
||
"dls_safety": 1,
|
||
"fb_clear": 0.0,
|
||
"fb_cmin": 0.3,
|
||
"sub_refresh": 0,
|
||
"fb_block_gain": 0.5,
|
||
"dump": "ml/runs/round52_ctl_ex_rows.json",
|
||
"out": "docs/experiment_round52_ctl_ex.json",
|
||
"trace": null,
|
||
"select": "endpoint",
|
||
"replan": "empty",
|
||
"safety": 1,
|
||
"partial": 1,
|
||
"inc_cap": 1,
|
||
"gap_margin": 0.0,
|
||
"cert_batch": 1,
|
||
"steps": 1,
|
||
"z_scale": 1.0,
|
||
"len_w": 0.0,
|
||
"tie": 0,
|
||
"pre_h": 0,
|
||
"pre_near": 0.0,
|
||
"backoff": 0,
|
||
"max_sec": null,
|
||
"seed": 0,
|
||
"dls": "analytic",
|
||
"device": "cuda"
|
||
},
|
||
"batch_size": 256,
|
||
"batch_seed_rule": "same seed per batch",
|
||
"checkpoint_sha256": "f7f686fd8ee92fb6948fbcc4ee954636065ad87bb54f661c4f8f575bc16c2ce1",
|
||
"tasks_sha256": "249bd93457825cda98c59e64d9a096d3473c69aeb3553de1dea6d9e56be71772",
|
||
"kin_fp": "2ebcbf17",
|
||
"act_scale": 1.0,
|
||
"metrics": {
|
||
"n": 256,
|
||
"n_ref": 256,
|
||
"success": 0.5234375,
|
||
"collision_free_success": 0.5234375,
|
||
"time_ok_1p3": 0.265625,
|
||
"collision_rate": 0.0,
|
||
"collision_steps_mean": 0.0,
|
||
"travel_ratio_median": 2.615795850753784,
|
||
"time_mean_s": 2.486567258834839,
|
||
"brakes_mean": 19.5546875,
|
||
"start_fail": 0.0,
|
||
"goal_ratio_2s": 0.27920001745224,
|
||
"err_fail_med": 204.71360778808594,
|
||
"err_success_med": 0.21751396358013153,
|
||
"dls_share": 0.61328125,
|
||
"time_budget": "teacher×2(≥3 s)",
|
||
"speed_ratio_p50": 1.0000016689300537,
|
||
"speed_ratio_p95": 1.0000101327896118,
|
||
"speed_ratio_max": 1.0000101327896118,
|
||
"over_vmax_rate": 0.0,
|
||
"accel_ratio_p95": 1.0000016689300537,
|
||
"accel_ratio_max": 1.0000101327896118,
|
||
"reverse_p50": 0.24741201102733612,
|
||
"reverse_mean": 0.25991496443748474,
|
||
"reverse_moving_p50": 0.32581019401550293,
|
||
"moving_share_p50": 0.9587156176567078,
|
||
"brake_rate_mean": 0.10028044879436493,
|
||
"brake_rate_p50": 0.0051282052882015705,
|
||
"brake_rate_p90": 0.3153846263885498,
|
||
"stuck_share": 0.234375,
|
||
"n_plans_mean": 195.0,
|
||
"success_reachable": 0.5514403581619263,
|
||
"by_cat": {
|
||
"A": [
|
||
128,
|
||
96
|
||
],
|
||
"B": [
|
||
74,
|
||
20
|
||
],
|
||
"C": [
|
||
32,
|
||
11
|
||
],
|
||
"D": [
|
||
13,
|
||
0
|
||
],
|
||
"Dr": [
|
||
9,
|
||
7
|
||
]
|
||
}
|
||
},
|
||
"simulation_wall_s": 296.3142095999938
|
||
}
|