summary stats across columns, where column names indicate groups

问题

Data frame have includes a few thousand vectors that follow a naming pattern. Each vector name includes a noun, then either _a, _b, or _c. Below are the first 10 vars and obs:

id  turtle_a   banana_a   castle_a   turtle_b   banana_b   castle_b   turtle_c   banana_c   castle_c
A      -0.58      -0.88      -0.56      -0.53      -0.32      -0.42      -0.52      -0.89      -0.72
B         NA         NA         NA      -0.84      -0.36      -0.26         NA         NA         NA
C       0.00      -0.43      -0.75      -0.35      -0.88      -0.14      -0.26      -0.15      -0.81
D      -0.81      -0.63      -0.77      -0.82      -0.83      -0.50      -0.77      -0.25      -0.07
E      -0.25      -0.33      -0.09      -0.51      -0.27      -0.81      -0.06      -0.23      -0.97
F      -0.80      -0.88      -0.05         NA         NA         NA         NA         NA         NA
G      -0.25      -0.76      -0.21         NA         NA         NA         NA         NA         NA
H      -0.47      -0.10      -0.67      -0.46      -0.71      -0.24      -0.76      -0.04      -0.11
I      -0.15      -0.34      -0.57      -0.40      -0.14      -0.49         NA         NA         NA
J      -0.65      -0.86      -0.37      -0.67      -0.81      -0.63         NA         NA         NA

Data frame want is the mean across all columns for every set of variables in a noun group. For example, averaging turtle_a, turtle_b, and turtle_c for id=A equals -0.54. Here's what want looks like if I just do it for the handful of noun groups in the example.

id   turtle_m    banana_m    castle_m
A       -0.54       -0.70       -0.57
B       -0.84       -0.36       -0.26
C       -0.20       -0.49       -0.57
D       -0.80       -0.57       -0.45
E       -0.27       -0.28       -0.62
F       -0.80       -0.88       -0.05
G       -0.25       -0.76       -0.21
H       -0.56       -0.29       -0.34
I       -0.27       -0.24       -0.53
J       -0.66       -0.83       -0.50

Options so far:

convert to long, summarize with a group_by() function in dplyr, and transpose back to wide.
resort the vectors so the noun groups appear next to each other, and write a loop that computes means across columns, taking three-column steps at each iteration

It seems like summarize_at or summarize_all could be used more effectively than either of my current options, but I'm not sure how to use it in a way that will dynamically group variables by naming convention.

Any thoughts?

回答1:

We can use split.default to split the columns based on the substring of column names, loop over the list with sapply with rowMeans and then cbind with the first column

out <- cbind(df1[1], sapply(split.default(df1[-1], 
    sub("_.*", "", names(df1)[-1])), rowMeans, na.rm = TRUE))

Or we can use pivot_longer

library(dplyr)
library(tidyr)
df1 %>% 
   pivot_longer(cols = -id, names_sep="_", names_to = c(".value", "group")) %>%
   group_by(id) %>%
   summarise(across(turtle:castle,  mean,  na.rm = TRUE))

来源：https://stackoverflow.com/questions/62722119/summary-stats-across-columns-where-column-names-indicate-groups

标签

loops

dplyr

data-manipulation

summary