R语言使用DKW不等式构建均匀置信带

DKW不等式指出，对于任何累积分布函数 $F$ 和所有自然数 $n$ ，有：
$$P(\underset{x}{\sup }|{\hat{F}}_n(x)-F(x)|>{\epsilon }_n)\le 2e^{-2n{\epsilon }_n^2}$$
为了使不等式的右边小于或等于 $\alpha$ ， ${\epsilon }_n$ 被选为：
$${\epsilon }_n=\sqrt{\frac{1}{2n}\ln \left(\frac{2}{\alpha }\right)}$$

# 使用R内置的faithful数据集
data("faithful")

# 提取等待时间数据
waiting_times <- faithful$waiting

# 计算经验分布函数
ecdf_data <- ecdf(waiting_times)

# 使用DKW不等式计算统一置信带
alpha <- 0.05
epsilon <- sqrt(log(2/alpha) / (2 * length(waiting_times)))

# 绘制ECDF及其置信带
x_vals <- seq(min(waiting_times), max(waiting_times), length.out = 100)
plot(ecdf_data, main = "Uniform Confidence Band", verticals = TRUE, do.points = FALSE)
lines(x_vals, pmax(0, ecdf_data(x_vals) - epsilon), col = "blue", lty = "dashed")
lines(x_vals, pmin(1, ecdf_data(x_vals) + epsilon), col = "blue", lty = "dashed")