如何在R中矢量化坐标计算的for循环？

Question

我正在尝试查看中每个 center点中有多少个center，但这是目前在for循环中完成的。如果可以矢量化的话会有可能吗？下面是一个可以复制的片段。谢谢。

require(geosphere)

centers <- as.data.frame(matrix(rnorm(10, mean = 40, sd = .5), ncol = 2, byrow = TRUE))
points <- matrix(rnorm(100, mean = 40, sd = 1), ncol = 2, byrow = TRUE)

for(i in 1:dim(centers)[1]){
  # Calculate number of points that are 50 km within every center point
  centers[i,3] <- sum(geosphere::distHaversine(points, 
                                               centers[i,c(1:2)]) /
                        1000 < 50, na.rm = TRUE)
}

Answer 1

您可以将split与row和sapply一起使用，然后再使用colSums

library(geosphere)
centers$res <- colSums(
  sapply(split(as.matrix(centers[, 1:2]), row(centers)[, 1:2]), 
         distHaversine, p1 = points) / 1000 < 50, na.rm = TRUE)

它给出了同样的结论：

# compute the old result to compare with
for(i in 1:dim(centers)[1])
  centers[i,4] <- sum(geosphere::distHaversine(points, 
                                               centers[i,c(1:2)]) /
                        1000 < 50, na.rm = TRUE)

# gives the same
all.equal(centers$res, centers[, 4])
#R> [1] TRUE

另一种可能的选择是：

dists <- tapply(as.matrix(centers[, 1:2]), row(centers[, 1:2]), 
                distHaversine, p1 = points)
centers$res <- colSums(simplify2array(dists) / 1000 < 50, na.rm = TRUE)

或者使用匿名函数。这就像Ronak Shah的答案，但是有了tapply

centers$res <- c(tapply(
  as.matrix(centers[, 1:2]), row(centers[, 1:2]), function(x)
    sum(distHaversine(points, x) / 1000 < 50, na.rm = TRUE)))

Answer 2

如果函数一次只能处理一个点，我不认为它能够真正地向量化。您可以将for循环替换为sapply，并查看性能是否有任何改善。

library(geosphere)

centers$total <- sapply(seq(nrow(centers)), function(i) {
      sum(distHaversine(points, centers[i,]) /1000 < 50, na.rm = TRUE)
})