cbb
af856ca728
两 seed 合并后必须更正轮次49 的一句话：到达题组内末态误差 vs 控制的符号检验 s0 67/57(p=0.419)、
s1 74/51(p=0.049)、合并 141差/108好 p=0.042 ⇒ 是"大幅减小"而不是"与控制不可分"。
中位数(+3.4%/+2.6%)与逐题分布给出相反印象，因为门控把尾部拉近中位、却把大量题各推一点。
到达两 seed 合并赢24丢21 净+3 p=0.766 ⇒ 这条轴上到达没有可用信号。
稳的部分：goal_ratio_2s 两 seed 都改善(0.170/0.201 vs 0.279/0.275) ⇒ "打分地平线=执行地平线"
买路上进度是唯一稳健效应。可推荐形态是 ungated pre，门控留给末态余量优先的场景。
统计自纠：精确 McNemar/符号检验原写法 2*Σ_{i<=min(k,n-k)}C(n,i)/2^n 在 k=n/2 中间点会算出 1.168>1；
改成 p=min(1,2*min(P(X<=k),P(X>=k))) 后重算本轮全部已上报 p 值，除中间点那例外数值不变
(0.125/0.424/0.678/0.455/0.327/0.042/4.9e-16/1.8e-12/2.4e-6 都对得上)，中间点正确值 p=1.0。
下一轮补 2×2 的缺失格 K16+pre 量可加性（已在盘三格），判读规则预注册。
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/round50_gate20_s1_rows.json",
|
||
"out": "docs/experiment_round50_gate20_s1.json",
|
||
"trace": null,
|
||
"select": "pre",
|
||
"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": 20.0,
|
||
"backoff": 0,
|
||
"max_sec": null,
|
||
"seed": 1,
|
||
"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": 0.53125,
|
||
"collision_free_success": 0.53125,
|
||
"time_ok_1p3": 0.30078125,
|
||
"collision_rate": 0.0,
|
||
"collision_steps_mean": 0.0,
|
||
"travel_ratio_median": 2.6468124389648438,
|
||
"time_mean_s": 2.462352752685547,
|
||
"brakes_mean": 19.71875,
|
||
"start_fail": 0.0,
|
||
"goal_ratio_2s": 0.20109058916568756,
|
||
"err_fail_med": 177.47691345214844,
|
||
"err_success_med": 0.22864246368408203,
|
||
"dls_share": 0.6484375,
|
||
"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": 0.2363184094429016,
|
||
"reverse_mean": 0.24333104491233826,
|
||
"reverse_moving_p50": 0.28763440251350403,
|
||
"moving_share_p50": 0.9736841917037964,
|
||
"brake_rate_mean": 0.10112179815769196,
|
||
"brake_rate_p50": 0.0051282052882015705,
|
||
"brake_rate_p90": 0.35641026496887207,
|
||
"stuck_share": 0.24609375,
|
||
"n_plans_mean": 195.0,
|
||
"success_reachable": 0.5596708059310913,
|
||
"by_cat": {
|
||
"A": [
|
||
128,
|
||
94
|
||
],
|
||
"B": [
|
||
74,
|
||
23
|
||
],
|
||
"C": [
|
||
32,
|
||
12
|
||
],
|
||
"D": [
|
||
13,
|
||
0
|
||
],
|
||
"Dr": [
|
||
9,
|
||
7
|
||
]
|
||
}
|
||
},
|
||
"simulation_wall_s": 271.2116083000001
|
||
}
|