Weird group_by + mutate + which.max behavior

Question

I'm running into unexpected behavior with dplyr:

library(dplyr)

df <- structure(list(date = c("2016-05-02", "2016-05-02", "2016-05-02", 
      "2016-05-02", "2016-05-02", "2016-05-02", "2016-05-02", "2016-05-02", 
      "2016-05-02", "2016-05-02", "2016-05-02", "2016-05-02", "2016-05-02", 
      "2016-05-02", "2016-05-02", "2016-05-02", "2016-05-02", "2016-05-02", 
      "2016-05-02", "2016-05-02", "2016-05-02", "2016-05-02", "2016-05-02", 
      "2016-05-02", "2016-05-02", "2016-05-02", "2016-05-02", "2016-05-02", 
      "2016-05-02", "2016-05-02", "2016-05-02", "2016-05-02", "2016-05-02", 
      "2016-05-02", "2016-05-02", "2016-05-02", "2016-05-02", "2016-05-02", 
      "2016-05-02", "2016-05-02", "2016-05-02", "2016-05-02", "2016-05-02", 
      "2016-05-02", "2016-05-02", "2016-05-02", "2016-05-02", "2016-05-02", 
      "2016-05-03", "2016-05-03", "2016-05-03", "2016-05-03", "2016-05-03", 
      "2016-05-03", "2016-05-03", "2016-05-03", "2016-05-03", "2016-05-03", 
      "2016-05-03", "2016-05-03", "2016-05-03", "2016-05-03", "2016-05-03", 
      "2016-05-03", "2016-05-03", "2016-05-03", "2016-05-03", "2016-05-03", 
      "2016-05-03", "2016-05-03", "2016-05-03", "2016-05-03", "2016-05-03", 
      "2016-05-03", "2016-05-03", "2016-05-03", "2016-05-03", "2016-05-03", 
      "2016-05-03", "2016-05-03", "2016-05-03", "2016-05-03", "2016-05-03", 
      "2016-05-03", "2016-05-03", "2016-05-03", "2016-05-03", "2016-05-03", 
      "2016-05-03", "2016-05-03", "2016-05-03", "2016-05-03", "2016-05-03", 
      "2016-05-03", "2016-05-03", "2016-05-03", "2016-05-04", "2016-05-04", 
      "2016-05-04", "2016-05-04", "2016-05-04", "2016-05-04", "2016-05-04", 
      "2016-05-04", "2016-05-04", "2016-05-04", "2016-05-04", "2016-05-04", 
      "2016-05-04", "2016-05-04", "2016-05-04", "2016-05-04", "2016-05-04", 
      "2016-05-04", "2016-05-04", "2016-05-04", "2016-05-04", "2016-05-04", 
      "2016-05-04", "2016-05-04", "2016-05-04", "2016-05-04", "2016-05-04", 
      "2016-05-04", "2016-05-04", "2016-05-04", "2016-05-04", "2016-05-04", 
      "2016-05-04", "2016-05-04", "2016-05-04", "2016-05-04", "2016-05-04", 
      "2016-05-04", "2016-05-04", "2016-05-04", "2016-05-04", "2016-05-04", 
      "2016-05-04", "2016-05-04", "2016-05-04", "2016-05-04", "2016-05-04", 
      "2016-05-04", "2016-05-05", "2016-05-05", "2016-05-05", "2016-05-05", 
      "2016-05-05", "2016-05-05", "2016-05-05", "2016-05-05", "2016-05-05", 
      "2016-05-05", "2016-05-05", "2016-05-05", "2016-05-05", "2016-05-05", 
      "2016-05-05", "2016-05-05", "2016-05-05", "2016-05-05", "2016-05-05", 
      "2016-05-05", "2016-05-05", "2016-05-05", "2016-05-05", "2016-05-05", 
      "2016-05-05", "2016-05-05", "2016-05-05", "2016-05-05", "2016-05-05", 
      "2016-05-05", "2016-05-05", "2016-05-05", "2016-05-05", "2016-05-05", 
      "2016-05-05", "2016-05-05", "2016-05-05", "2016-05-05", "2016-05-05", 
      "2016-05-05", "2016-05-05", "2016-05-05", "2016-05-05", "2016-05-05", 
      "2016-05-05", "2016-05-05", "2016-05-05", "2016-05-05", "2016-05-06", 
      "2016-05-06", "2016-05-06", "2016-05-06", "2016-05-06", "2016-05-06", 
      "2016-05-06", "2016-05-06", "2016-05-06", "2016-05-06", "2016-05-06", 
      "2016-05-06", "2016-05-06", "2016-05-06", "2016-05-06", "2016-05-06", 
      "2016-05-06", "2016-05-06", "2016-05-06", "2016-05-06", "2016-05-06", 
      "2016-05-06", "2016-05-06", "2016-05-06", "2016-05-06", "2016-05-06", 
      "2016-05-06", "2016-05-06", "2016-05-06", "2016-05-06", "2016-05-06", 
      "2016-05-06", "2016-05-06", "2016-05-06", "2016-05-06", "2016-05-06", 
      "2016-05-06", "2016-05-06", "2016-05-06", "2016-05-06", "2016-05-06", 
      "2016-05-06", "2016-05-06", "2016-05-06", "2016-05-06", "2016-05-06", 
      "2016-05-06", "2016-05-06"), abc = c(NA, NA, NA, NA, NA, NA, 
         NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, 
         NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, 
         NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, 0, 0, 0, 0, 0, 0, 0, 
         0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 10, 20, 20, 16, 
         14, 9, 8, 6, 5, 5, 6, 7, 13, 24, 52, 65, 68, 66, 65, 58, 47, 
         21, 6, 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 
         1, 1, 0, 0, 0, 0, 0, 10, 19, 19, 15, 11, 8, 8, 5, 4, 4, 4, 5, 
         9, 17, 31, 43, 49, 52, 52, 47, 32, 21, 6, 2, 1, 1, 1, 1, 1, 1, 
         1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 5, 14, 
         14, 14, 15, 18, 18, 14, 14, 14, 15, 19, 29, 46, 58, 62, 69, 71, 
         67, 56, 40, 25, 8, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 
         2, 2, 2, 2, 2, 2, 0, 0, 0, 0, 0, 10, 18, 18, 14, 12, 9, 7, 5, 
         4, 5, 5, 7, 9, 17, 30, 36, 49, 52, 54, 54, 42, 32, 15, 5, 1)), 
     class = "data.frame", row.names = c(NA, -240L), .Names = c("date", "abc"))


df %>%
  group_by(date) %>%
  mutate(peak_max_index = as.numeric(which.max(as.numeric(abc))))

What I would expect this to return is peak_max_index that's 41 for all rows where date is 2016-05-04. But strangely peak_max_index is NA instead. Even more strangely, if you kick out all rows where date is 2016-05-03 before you run the dplyr commands, the result is entirely correct. Is this a bug?

Did you try df %>% group_by(date) %>% mutate(peak_max_index = as.numeric(which.max(as.numeric(abc)))) %>% filter(date == '2016-05-04')? This is showing that the first part is doing the right thing. What is packageVersion('dplyr') show? — Gopala
So, what part of the result is weird? I saved the result of your command into df and subset this way df[df$date == '2016-05-04', ] and still get 41 for all rows. — Gopala
BTW - there are some bugs in dplyr 0.4.3 (unrelated to this issue) and so I use dev version 0.4.3.9001. — Gopala
For the same subset, I get `NA' for all rows. May have to try the dev version. — RoyalTS

seasmith seasmith · Accepted Answer · 2016-05-03T21:49:31

You are evaluating NA's in your formula which.max(). Just eliminate the NA's with !is.na().

df %>%
    group_by(date) %>%
    mutate(peak_max_index = max(df$abc[!is.na(df$abc)]))

Weird group_by + mutate + which.max behavior

1 Answers