假阳性与假阴性权衡图

Question

我正在进行决策分析，我试图使用 R 来说明假阳性（假通过）与假阴性（假不通过）之间的权衡。我创建了一个密度图具有零假设曲线和替代假设曲线，但想通过这些示例图进一步说明这种关系。感谢您帮助创建示例 1 和示例 2 等图，尤其是示例 1。谢谢！

密度图

示例 1

示例 2

Answer 1

看来您的症结在于如何在给定的 x 或 y 值处访问密度曲线的值。

您可以使用 ggplot_build() 访问由 geom_density 函数构造的底层 data.frame。 Here is some further discussion.

Answer 2

我使用了两个偏移正态分布，前提是你可以想象任何曲线。

library(ggplot2)

xs <- seq(-2, 4, length.out = 201)
dat <- do.call(rbind,
  list(data.frame(x=xs, y=dnorm(xs), id="1"),
       data.frame(x=xs, y=dnorm(xs, 2), id="2")))

地块 1

vline <- 1
eps <- 1e-3

ggplot(dat, aes(x, y, group = id, color = id)) +
  geom_line() +
  geom_area(aes(fill = id),
            data = ~ subset(., (id == "1" & x > (vline+eps)) | (id == "2" & x < (vline-eps)))) +
  geom_vline(xintercept = vline, linetype = "dashed") +
  labs(x = "Hazard Ratio", y = NULL) +
  guides(color = "none", fill = "none") +
  theme_classic() +
  theme(
    axis.line.y = element_blank(),
    axis.text.y = element_blank(),
    axis.ticks.y = element_blank()
  )

这里的vline是分界线，不在路口的话还是有用的。例如，

vline <- 1.2

地块 2

rng <- c(0.75, 0.85)
rngdat <- do.call(rbind,
  by(dat, dat$id, function(z) with(z, data.frame(approx(x, y, xout = rng), id = id[1]))))
rngdat$otherx <- fifelse(rngdat$id == "1", Inf, -Inf)

ggplot(dat, aes(x, y, group = id, color = id)) +
  geom_line(na.rm = TRUE) +
  geom_segment(aes(xend = x, yend = 0),
               data = subset(rngdat, id == 1),
               color = "black", linetype = "dashed") +
  geom_segment(aes(xend = otherx, yend = y),
               data = rngdat, linetype = "dashed") +
  coord_cartesian(xlim = c(0, 2)) +
  scale_x_continuous(name = "HR gate") +
  scale_y_continuous(
    name = "False Go Probability",
    sec.axis = sec_axis(~ ., name = "False No-Go Probability")) +
  scale_color_manual(values = c("1" = "blue", "2" = "red")) +
  guides(color = "none") +
  theme_classic() +
  theme(
    axis.line.y.left = element_line(color = "red"),
    axis.line.y.right = element_line(color = "blue")
  )

情节 3

offset <- max(rngdat$y[rngdat$id == "1"]) + 0.1
cutoff <- 0
dat <- transform(
  dat,
  yoff = ifelse(id == "1", 0.05 + offset, 0),
  cat = ifelse(id == "1",
               ifelse(x < cutoff, "True Positive", "False Negative"),
               ifelse(x < cutoff, "False Positive", "True Negative")))

ggplot(dat, aes(x, y = y + yoff)) +
  geom_ribbon(aes(ymin = yoff, ymax = y + yoff,
                  group = cat, fill = cat, alpha = cat),
              na.rm = TRUE) +
  geom_vline(xintercept = cutoff) +
  scale_fill_manual(
    name = NULL,
    values = c("True Positive" = "red", "False Negative" = "red",
               "False Positive" = "blue", "True Negative" = "blue")) +
  scale_alpha_manual(
    name = NULL,
    values = c("True Positive" = 1, "False Negative" = 0.2,
               "False Positive" = 0.2, "True Negative" = 1)) +
  labs(x = NULL, y = NULL) +
  theme(
    legend.position = "bottom",
    axis.text.y = element_blank(),
    axis.ticks.y = element_blank()
  )

假阳性与假阴性权衡图

False positive vs. false negative trade off plot

r

false-positive

ggplot2

地块 1

地块 2

情节 3