在考虑值的类型的情况下循环数据帧

Question

我有一个大的坐标数据集，我想计算它们之间的欧几里得距离。示例如下：

df <- data.frame("name" = c("a","b","c","a","e"), "type" = c("me","me","me","we", "we"), 
    "x" = c(64.044,63.722,64.359,65.373, 65.122),"y" = c(51.615,52.849,53.119,51.805,52.78),
    "z" = c(33.423,32.671,31.662,34.158,35.26))

name  type    x   y   z
a me  64.044  51.615  33.423
b me  63.722  52.849  32.671
c me  64.359  53.119  31.662
d we  65.373  51.805  34.158
e we  65.122  52.78   35.26

我想计算具有不同类型的名称列值的行之间的欧几里得距离值。我写了这段代码

require("distances")

for (i in 1:nrow(df)) {
  if(!(df$type[i]%in%df$type[i+1])){
    d <- distances(df[,3:5])
  }
}

然而这是错误的。我的理想结果应该如下所示，其中 d 说明了名称值之间的距离：

    a.me    b.me    c.me    a.we    e.we
a.me    0   0   0   d   d
b.me    0   0   0   d   d
c.me    0   0   0   d   d
a.we    d   d   d   0   0
e.we    d   d   d   0   0

此外，我的数据集变化很大，我认为 for 不是最佳选择。任何人都可以在这个问题上帮助我吗？

Answer 1

这是一个解决方案，但根据您的数据大小可能需要大量计算工作，因为首先计算整个 dist-matrix，然后只选择需要的那些：

dist_mat <- dist(df[3:5], diag = TRUE, upper = TRUE)
dist_mat <- as.matrix(dist_mat)
names_vec <- paste(df$name, df$type, sep = "_")
dimnames(dist_mat) <- list(names_vec, names_vec)

dist_mat <- 
  sapply(colnames(dist_mat), 
         function(x) ifelse(grepl(strsplit(x, "_")[[1]][2], 
                                  row.names(dist_mat)), NA, dist_mat[,x]))

row.names(dist_mat) <- names_vec

dist_mat
#          a_me     b_me     c_me     d_we     e_we
# a_me       NA       NA       NA 1.530544 2.427731
# b_me       NA       NA       NA 2.454976 2.944093
# c_me       NA       NA       NA 2.997467 3.693602
# d_we 1.530544 2.454976 2.997467       NA       NA
# e_we 2.427731 2.944093 3.693602       NA       NA

Answer 2

这就是诀窍：

  df <- data.frame("name" = c("a","b","c","d","e"), "type" = c("me","me","me","we", "we"), 
                   "x" = c(64.044,63.722,64.359,65.373, 65.122),"y" = c(51.615,52.849,53.119,51.805,52.78),
                   "z" = c(33.423,32.671,31.662,34.158,35.26))

# lapply over the levels of the type
x <- lapply(levels(df$type), function(level) {
  # select the matching rows and columns and convert
  mat <- as.matrix(df[as.character(df$type) == level, 3:5])
  # names are set as row names for dist to use
  row.names(mat) <- paste(df$name[as.character(df$type) == level],
                          level,
                          sep = "_")
  # measuring the distance
  dist <- dist(mat, method = "euclidean", diag = TRUE, upper = TRUE)
  # converting distance to matrix
  as.matrix(dist)
})
# bind the list to one matrix
x <- plyr::rbind.fill.matrix(x)
# add rownames
row.names(x) <- colnames(x)
x

a_me     b_me     c_me     d_we     e_we
a_me 0.000000 1.480522 2.337170       NA       NA
b_me 1.480522 0.000000 1.223417       NA       NA
c_me 2.337170 1.223417 0.000000       NA       NA
d_we       NA       NA       NA 0.000000 1.492659
e_we       NA       NA       NA 1.492659 0.000000