How to order rows in data frame for juts maximun net values and symbols for some columns in R-CodePudding

I have this dataframe

a <- c("5*", 7, 9, "11***")
b <- c("-8*", "-10*", -3, -1)
c <- c(-4, -1, "-6**", "3*")
df <- data.frame(a,b,c)

    a        b       c
1   5*      -8*     -4
2   7       -10*    -1
3   9       -3      -6**
4   11***   -1       3*

then I want the rows be ordered for the maximum net value in them (just considering those values with at least one "*"), Independently of the columns. I mean, I expect something like this:

    a        b       c
4   11***   -1       3*
2   7       -10*    -1
1   5*      -8*     -4
3   9       -3      -6**

as you can see it was sorted because the first row has the value of 11***, the second (-10*), the third (-8*), the fourth (-6**). You can notice that in the last row it was omited the 9 in the sorting, that's because we just want to consider those numbers with at least one "*"

CodePudding user response：

One way - get the absolute max among the * suffixed values, create a order index based on those and use that to reorder the rows

df[order(-apply(df, 1, FUN = function(x) 
       max(abs(as.numeric(sub("\\* ", "", x[grep("\\*", x)] )))))),]

-output

    a    b    c
4 11***   -1   3*
2     7 -10*   -1
1    5*  -8*   -4
3     9   -3 -6**

or using tidyverse, reshape to 'long' format with pivot_longer, filter only the elements having *, arrange on the numeric part by extraction (parse_number), slice the rows of original based on the row number arranged

library(dplyr)
library(stringr)
library(tidyr)
df %>%
    mutate(rn = row_number()) %>% 
    pivot_longer(cols = -rn) %>%
    filter(str_detect(value, fixed("*"))) %>%
    arrange(desc(abs(readr::parse_number(value)))) %>%
    distinct(rn) %>%
    pull(rn) %>%
    slice(df, .)

-output

     a    b    c
1 11***   -1   3*
2     7 -10*   -1
3    5*  -8*   -4
4     9   -3 -6**