关于「视锥剔除:与现金流没有因果路径的任务」,《瓶颈定焦与价值工时比排程》给出的定量结论是什么?
结论:完全不误判时,剔除让每天的价值从 110.7 点升到 219.1 点,多出 0.979,接近翻倍。误判率 0.1 时多出 0.846,误判率 0.3 时仍多出 0.548,误判率高到 0.5,也就是每两件看错一件,剔除仍然多出 0.200。代价写在最后一列:留在挂起区里的价值随误判率上升,从 17.8 点升到 155.7 点。
数据口径:史密斯 1956 年证明,单人面对一张清单,使加权延迟代价最小的次序是按价值除以耗时排序,而不是按价值排序;七件示意任务上,按重要性排比穷举最优多付 0.311。 图形学的视锥剔除与细节层次说明,固定帧预算下不该每件都精做:八件示意任务全部精做只做完三件,总价值只有最优安排的 0.783。
对照数据表
下表数字由本章 Python 程序原样打印,可重新运行复现。 程序只用 Python 标准库与结论吻合的打印值: 0.1, 0.200, 0.3, 0.5, 0.548, 0.846
| e | list | cull | gain | value left in the cull pile |
|---|---|---|---|---|
| 0.0 | 110.7 | 219.1 | 0.979 | 17.8 |
| 0.1 | 110.7 | 204.3 | 0.846 | 45.8 |
| 0.2 | 110.7 | 188.8 | 0.706 | 73.0 |
| 0.3 | 110.7 | 171.4 | 0.548 | 100.5 |
| 0.5 | 110.7 | 132.8 | 0.200 | 155.7 |
查看完整验算程序
# Chapter 4. p42: what culling costs when the task graph is misjudged.
# Each simulated day has 12 candidate tasks of 20 to 120 minutes and a
# focus budget of 300 minutes. True class: IN with prob 0.25 (value 1.0
# per minute), DEFER 0.40 (0.25 per minute), CULL 0.35 (0, except that
# one in ten hides an unseen path worth 0.6 per minute).
# The operator misreads each task with prob e: an IN or DEFER task is
# culled, or a CULL task is kept as DEFER. Policies:
# list work the list in random order until the budget is spent
# cull IN first, then DEFER, never CULL (by the operator's labels)
# 10,000 days per row, common random numbers. Illustrative values.
import math
import random
import statistics
RATE = {"IN": 1.0, "DEFER": 0.25, "CULL": 0.0}
def make_day(rng):
day = []
for _ in range(12):
m = rng.choice(range(20, 121, 10))
u = rng.random()
c = "IN" if u < 0.25 else ("DEFER" if u < 0.65 else "CULL")
hidden = c == "CULL" and rng.random() < 0.1
day.append((m, c, 0.6 if hidden else RATE[c]))
return day
def spend(order, budget=300):
left, val = budget, []
for m, rate in order:
use = min(m, left)
val.append(use * rate)
left -= use
if left <= 0:
break
return math.fsum(val)
print("e list cull gain value left in the cull pile")
for e in (0.0, 0.1, 0.2, 0.3, 0.5):
rng = random.Random(37642)
a, b, lost = [], [], []
for _ in range(10000):
day = make_day(rng)
shuffled = day[:]
rng.shuffle(shuffled)
a.append(spend([(m, r) for m, _, r in shuffled]))
lab = []
for m, c, r in day:
if rng.random() < e:
c2 = "CULL" if c != "CULL" else "DEFER"
else:
c2 = c
lab.append((m, c2, r))
ins = [(m, r) for m, c2, r in lab if c2 == "IN"]
dfr = [(m, r) for m, c2, r in lab if c2 == "DEFER"]
b.append(spend(ins + dfr))
lost.append(math.fsum(m * r for m, c2, r in lab if c2 == "CULL"))
ma, mb = statistics.fmean(a), statistics.fmean(b)
print("%.1f %6.1f %6.1f %6.3f %6.1f" % (
e, ma, mb, mb / ma - 1.0, statistics.fmean(lost)))唯一数据源与引用
范雅福(Fan, Y.). (2026). 《阶段瓶颈动态定焦、价值工时比优先级排序与认知视锥渲染加速母典》第 4 章「视锥剔除:与现金流没有因果路径的任务」(版本 1.0.0). 5D Library. DOI 待发布
DOI 待发布 · 专著页正式发布后上线(本站暂无专著页) · 5DL-B-376 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026376,
author = {Fan, Yafu},
title = {Stage-Bottleneck Triage, Weighted Shortest-Processing-Time Scheduling and Level-of-Detail Execution Velocity},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}