Pfthb

I would like to summarize data across multiple columns, based on all rows other than rows with a certain value in a separate grouping variable column. For example, in the df below, I want to get the medians of A, B, C, D, and E based on values from rows that are not assigned to the cluster matching a given row.

df = data.frame(cluster = c(1:5, 1:3, 1:2),
 A = rnorm(10, 2),
 B = rnorm(10, 5),
 C = rnorm(10, 0.4),
 D = rnorm(10, 3),
 E = rnorm(10, 1))

df %>%
group_by(cluster) %>%
summarise_at(toupper(letters[1:5]), funs(m = fun_i_need_help_with(.)))

fun_i_need_help_with would give the equivalent of:

 first row: median(df[which(df$cluster != 1), "A"])
 second row: median(df[which(df$cluster != 2), "A"])
 and so on...

I can do it with nested for loops, but it's pretty slow to run and doesn't seem like a good R-like solution.

for(col in toupper(letters[1:5]))
 for(clust in unique(df$cluster))
 df[which(df$cluster == clust), col] <-
 median(df[which(df$cluster != clust), col])
 
 

A solution using the tidyverse.

set.seed(123)

df = data.frame(cluster = c(1:5, 1:3, 1:2),
 A = rnorm(10, 2),
 B = rnorm(10, 5),
 C = rnorm(10, 0.4),
 D = rnorm(10, 3),
 E = rnorm(10, 1))

library(tidyverse)

df2 <- map_dfr(unique(df$cluster),
 ~df %>%
 filter(cluster != .x) %>%
 summarize_at(vars(-cluster), funs(median(.))) %>%
 # Add a label to show the content of this row is not from a certain cluster number
 mutate(not_cluster = .x))
df2
# A B C D E not_cluster
# 1 2.070508 5.110683 0.1820251 3.553918 0.7920827 1
# 2 2.070508 5.400771 -0.6260044 3.688640 0.5333446 2
# 3 1.920165 5.428832 -0.2769652 3.490191 0.8543568 3
# 4 1.769823 5.400771 -0.2250393 3.426464 0.5971152 4
# 5 1.769823 5.400771 -0.3288912 3.426464 0.5971152 5