子集中两组之间的最短欧式距离

Question

I have a largish data frame (50000 points) representing points in 2D collected from biological images. Points are categorised as either red or green and are associated with each other in groups (in the example: cells AD). A small test data set (MSR_test.csv) can be found here .

require(ggplot2)
cells <- read.csv("MSR_test.csv")
ggplot(cells, aes(X, Y, colour = channel, shape = cell)) + 
   geom_point() +
   scale_colour_manual(values = c("green","red"), name = "channel")

I am trying to find a reasonably straight forward way (perhaps involving plyr ?) to find the Euclidean distance between each green point and its nearest red point within the same 'cell group'. Whilst I think I have worked out how to do this for an individual grouping (using rdist from package fields ) I can't seem to work out how to apply a method to my data frame.

Answer 1

I don't see any reason to use plyr , but maybe I'm wrong. The following code works on your example. I did not use any heavy function to compute the Euclidean distance, mainly because you may have to compute it on a lot of points.

green <- subset(cells, channel=="Green")
red <- subset(cells, channel=="Red")
fun_dist <- function(a, M) rowSums( (M - matrix(1,nrow(M),1) %*% as.numeric(a))**2 )
foo <- function(greenrow, matred) {
  subred <- subset(matred, cell == greenrow["cell"], select=c("X","Y"))
  minred <- subred[ which.min(fun_dist(unlist(greenrow[c("X","Y")]),subred)), ]
  return(minred)
}
data.frame( "rbind", apply(green, 1, foo, red) )