关于「从流量序列读系数:共同波动、暂停试验与两两判别的准确率」,《站群竞争相图》给出的定量结论是什么?

结论:第六章说明被动观察几乎读不出系数,暂停试验能读出,两两结局预测三站结局的准确率为 0.934。

数据口径:立项命题经核对后修正:系数大于一时,只有被经营者撑住的内部平衡点低于单站,自由竞争的结局是只剩一个站;两个系数之积小于一只是共存的必要条件,示意网格中满足它的格子仍有 0.627 以一方出局告终。 书中站、账号、访客、成本与经营者均为示意;全部计算由书中程序生成;本书是非个人化经营建议,只讨论以真实主体运营、彼此关系公开的合规品牌矩阵,不涉及马甲账号与刷量。

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: 0.934

trials 400, rule gives a prediction in 287
no prediction (a bistable pair or no unique set)113
correct 268, wrong 19, accuracy 0.934
rule says all three persist10 cases, correct 7

核心公式

\[\ln\frac{N_{1,t+1}}{N_{1,t}} = r\left(1 - \frac{N_{1,t} + a_{12} N_{2,t}}{K}\right) + u_t + e_{1,t},\]
查看完整验算程序
# Chapter 6. p63: do pairwise verdicts predict a three-site matrix?
# Friedman, Higgins and Gore (2017) found that pairwise outcomes
# predicted which of three bacterial species survive about 90 percent
# of the time. Their rule: sites that coexist in every pair persist
# together; a site excluded by any survivor drops out. We test the
# rule on 400 random three-site matrices (illustrative ranges):
# K_i uniform on 5000 to 15000, a_ij uniform on 0.2 to 1.6, r = 0.3.
# Truth: Euler steps of 0.2 month for 1200 months from 1000 visitors
# each; a site survives if it ends above 1 percent of its K.
import math
import random
from itertools import combinations

rng = random.Random(20261115)
TRIALS = 400


def verdict(i, j, a, k):
    hold_i = a[j][i] > k[j] / k[i]      # i alone resists j
    hold_j = a[i][j] > k[i] / k[j]      # j alone resists i
    if hold_i and hold_j:
        return "bi"
    if hold_i:
        return i
    if hold_j:
        return j
    return "co"


def predict(a, k):
    v = {(i, j): verdict(i, j, a, k) for i, j in combinations(range(3), 2)}
    if "bi" in v.values():
        return None
    out = []
    for size in (3, 2, 1):
        for s in combinations(range(3), size):
            inside = all(v[p] == "co" for p in combinations(s, 2))
            beaten = all(any(v[tuple(sorted((x, y)))] == y for y in s)
                         for x in range(3) if x not in s)
            if inside and beaten:
                out.append(frozenset(s))
    return out[0] if len(out) == 1 else None


def truth(a, k):
    n = [1000.0, 1000.0, 1000.0]
    for _ in range(6000):
        n = [n[i] + 0.2 * 0.3 * n[i] * (1 - math.fsum(
             x * y for x, y in zip(n, a[i])) / k[i]) for i in range(3)]
    return frozenset(i for i in range(3) if n[i] > 0.01 * k[i])


hit = miss = nopred = 0
rich = rich_hit = 0
for _ in range(TRIALS):
    k = [rng.uniform(5000, 15000) for _ in range(3)]
    a = [[1.0 if i == j else rng.uniform(0.2, 1.6) for j in range(3)]
         for i in range(3)]
    p = predict(a, k)
    if p is None:
        nopred += 1
        continue
    t = truth(a, k)
    if p == t:
        hit += 1
    else:
        miss += 1
    if len(p) == 3:
        rich += 1
        rich_hit += p == t
print("trials %d, rule gives a prediction in %d" % (TRIALS, hit + miss))
print("no prediction (a bistable pair or no unique set): %d" % nopred)
print("correct %d, wrong %d, accuracy %.3f" % (hit, miss,
                                                hit / (hit + miss)))
print("rule says all three persist: %d cases, correct %d" % (rich,
                                                              rich_hit))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《洛特卡沃尔泰拉竞争相图、多子站多账号矩阵内耗与共生壁垒判定母典》第 6 章「从流量序列读系数:共同波动、暂停试验与两两判别的准确率」(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-251 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026251,
  author    = {Fan, Yafu},
  title     = {Lotka-Volterra Competition Phase Portraits: Cannibalisation within Multi-Site, Multi-Account Matrices versus Symbiotic Barriers},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论