根据多个向量的多个条件创建数字样本

我想我明白。您想要在由距离分隔的间隙中进行采样，而复杂的因素是您不能在标记距离的任一侧进行750m的采样。

我认为对问题进行更清晰的视觉理解很有用。在此图中，x轴表示距离（y轴只是一个“虚拟”轴，因为我们只对x轴感兴趣）。黑条是我们无法采样的“排除区域”。在我们不希望采样的禁区的两侧，也有750m区域，这里用红色表示：

因此，从本质上讲，我们希望从该图中的x轴非阴影区域获得均匀的样本。

我的解决方案是首先合并重叠的片段，然后创建一个根据每个间隙的大小加权的样本空间，并从该空间中获取50个均匀样本。

在这里，我已概括为允许任意限制和样本大小。

sample_space <- function(km_ini, km_fin, km_max = 99.45, buffer = 0.75, n = 50)
{
  # Find and merge overlaps
  i <- 1
  km_ini <- km_ini - buffer
  km_fin <- km_fin + buffer
  while(i <= length(km_ini))
  {
    overlaps <- which(km_ini < km_fin[i] & km_fin > km_ini[i])
    if(length(overlaps) < 2) {i <- i + 1; next;}
    km_ini <- c(km_ini, min(km_ini[overlaps]))
    km_fin <- c(km_fin, max(km_fin[overlaps]))
    km_ini <- km_ini[-overlaps]
    km_fin <- km_fin[-overlaps]
    i <- 1
  }

  # Create a matrix of appropriate gaps
  gaps <- cbind(sort(km_fin), c(sort(km_ini)[-1], km_max))

  print(gaps)
  # Create a weighted sample space
  splits <- c(0, cumsum(apply(gaps, 1, diff)))

  # Take a sample within that space
  runifs <- runif(n, 0, max(splits))

  # Add the appropriate starting value back on
  index <- as.numeric(cut(runifs, splits))
  runifs - splits[index] + gaps[index, 1]
}

所以现在我们可以做

sample_space(df$km_ini, df$km_fin)
#>  [1] 93.107858 92.216660 83.597703 86.341198 72.258245 86.591883 18.572744
#>  [8] 16.641163 73.344658 73.075426 78.230074 97.745802 52.654342 52.298444
#> [15] 70.029034 67.430346 95.328900 62.250864 79.144025 86.344868  7.063474
#> [22] 58.797335 79.304272 54.731057 32.137068 84.837576 94.226992 50.808135
#> [29] 65.987277 76.666933 29.225744 33.309866 13.013735  6.925277 65.207383
#> [36] 91.591293 61.614993 18.646062 97.550237 48.478875 12.860920 20.263471
#> [43] 34.980616 50.583291 15.813562 96.104448 91.310377 53.063613 17.376281
#> [50] 72.511153

为了表明这确实达到了我们想要的目的，让我们在排除区域的图上绘制样本：

set.seed(69)
sample_df <- data.frame(x = sample_space(df$km_ini, df$km_fin),
                      y = runif(50, 0.45, 0.55))

ggplot(df) + 
  geom_rect(aes(xmin = km_ini - 0.75, xmax = km_fin + 0.75, ymin = 0, ymax = 1), 
            alpha = 0.5, fill = "red") +
  geom_rect(aes(xmin = km_ini, xmax = km_fin, ymin = 0, ymax = 1), fill = "black") +
  geom_rect(aes(xmin = 0, xmax = 0.75, ymin = 0, ymax = 1), alpha = 0.5) +
  geom_rect(aes(xmin = 99.45, xmax = 100, ymin = 0, ymax = 1), alpha = 0.5) +
  labs(x = "distance", y = "dummy") +
  geom_point(data = sample_df, aes(x = x, y = y), col = "blue")

^{由reprex软件包（v0.3.0）创建于2020-03-01}