关于「从吞吐记录拟合 σ 与 κ:最小二乘、自助区间与观测设计」,《并发拓扑耦合与黑板复用》给出的定量结论是什么?

结论:第二章用线性化最小二乘拟合 σ 与 κ,分层自助法给出峰值区间 6.37 到 7.80,并显示八个窗口有 0.962 的重抽落在峰值右侧。

数据口径:书中参数为示意,数据为合成,全部计算由书中程序生成;本书是非个人化经营建议,不提供任何加速比承诺。 第一章说明峰值从何而来,示意参数下峰值在 6.78 个窗口附近。

English version

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: 0.962, 6.37, 7.80

A. 2,000 bootstrap refits, 0 without a finite peak
quantity5 % 50 % 95 %
sigma0.0627 0.1034 0.1483
kappa0.0141 0.0189 0.0229
N*6.37 6.89 7.80
peak X29.44 30.10 30.82
B. share of refits with N* below N6: 0.000
B. share of refits with N* below N8: 0.962
B. share of refits with N* below N10: 1.000
C. integer band for N*6 to 8
查看完整验算程序
# Chapter 2. p22: bootstrap intervals for sigma, kappa, N* and the peak.
# Same synthetic data as p21 (seed 41121). Resample observations with
# replacement inside each level N (stratified pairs bootstrap, Efron),
# refit, and read 5 % and 95 % percentiles. B = 2000 resamples.
import math
import random

TRUE_S, TRUE_K, X1 = 0.08, 0.02, 10.0
LEVELS = (1, 2, 3, 4, 6, 8, 10, 12)
REPS = 3
B = 2000


def usl(n, s, k):
    return n / (1.0 + s * (n - 1) + k * n * (n - 1))


def observe(rng):
    data = []
    for n in LEVELS:
        for _ in range(REPS):
            noise = math.exp(rng.gauss(0.0, 0.06))
            data.append((n, X1 * usl(n, TRUE_S, TRUE_K) * noise))
    return data


def fit(data):
    base = [x for n, x in data if n == 1]
    x1 = math.fsum(base) / len(base)
    rows = [(n - 1, n * (n - 1), n / (x / x1) - 1.0)
            for n, x in data if n > 1]
    a = math.fsum(r[0] * r[0] for r in rows)
    b = math.fsum(r[0] * r[1] for r in rows)
    d = math.fsum(r[1] * r[1] for r in rows)
    e = math.fsum(r[0] * r[2] for r in rows)
    f = math.fsum(r[1] * r[2] for r in rows)
    det = a * d - b * b
    return x1, (e * d - b * f) / det, (a * f - b * e) / det


def pct(xs, q):
    ys = sorted(xs)
    return ys[min(len(ys) - 1, int(q * len(ys)))]


if __name__ == "__main__":
    data = observe(random.Random(41121))
    groups = {n: [d for d in data if d[0] == n] for n in LEVELS}
    rng = random.Random(41122)
    ss, ks, ns, ps = [], [], [], []
    no_peak = 0
    for _ in range(B):
        boot = []
        for n in LEVELS:
            g = groups[n]
            boot += [g[rng.randrange(len(g))] for _ in g]
        x1, s, k = fit(boot)
        ss.append(s)
        ks.append(k)
        if k <= 0.0 or s >= 1.0:
            no_peak += 1
            continue
        nstar = math.sqrt((1.0 - s) / k)
        ns.append(nstar)
        ps.append(x1 * usl(nstar, s, k))
    print("A. {:,} bootstrap refits, {} without a finite peak".format(
        B, no_peak))
    print("   quantity      5 %      50 %     95 %")
    for name, xs, fmt in (("sigma", ss, "{:8.4f}"),
                          ("kappa", ks, "{:8.4f}"),
                          ("N*", ns, "{:8.2f}"),
                          ("peak X", ps, "{:8.2f}")):
        print("   {:<9} ".format(name) + " ".join(
            fmt.format(pct(xs, q)) for q in (0.05, 0.5, 0.95)))
    for cur in (6, 8, 10):
        past = 0
        for v in ns:
            if v < cur:
                past += 1
        print("B. share of refits with N* below N = {:>2}: {:.3f}".format(
            cur, past / len(ns)))
    lo, hi = pct(ns, 0.05), pct(ns, 0.95)
    print("C. integer band for N*: {} to {}".format(
        math.floor(lo), math.ceil(hi)))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《多核并发操作拓扑耦合、共享黑板无锁通信与跨节点涌现增益母典》第 2 章「从吞吐记录拟合 σ 与 κ:最小二乘、自助区间与观测设计」(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-411 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026411,
  author    = {Fan, Yafu},
  title     = {Topological Coupling of Concurrent Multi-Core Operations: Lock-Free Blackboard Signalling and Cross-Node Emergent Gains},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论