group_by() %>% mutate(factor()) strange behavior

[paulanka]

Using factor() in mutate on a grouped data frame gives strange results

require(Lahman)
d <- Batting %>% group_by(lgID,yearID) %>%
   summarise(s = sum(G)) %>%
   mutate(f0=s>9000,
               f1=factor(s>9000))
xtabs(~f0,d) # 20 250
xtabs(~f1,d) # 136 134 ( wrong)

using ungroup() before mutate() solves the problem

d <- Batting %>% group_by(lgID,yearID) %>%
  summarise(s = sum(G)) %>%
  ungroup() %>%
  mutate(f0=s>9000,
   f1=factor(s>9000))
xtabs(~f0,d) # 20 250
xtabs(~f1,d) # 20 250

[romainfrancois]

This is because the distribution of unique values is not the same across groups:

> fact <- function(x) { print(unique(x)); factor(x) }
>
>
> d <- Batting %>% group_by(lgID,yearID) %>%
+    summarise(s = sum(G)) %>%
+    mutate(f0=s>9000,f1= fact(s>9000) )
[1] FALSE  TRUE
[1] TRUE
[1] TRUE
[1] FALSE
[1] FALSE  TRUE
[1] TRUE
[1] FALSE

[paulanka]

Thank you for your quick reply. This is a bit tricky. Maybe the doc should warn about using factor with grouped df.

[romainfrancois]

Nah that's a bug. I'll probably pick it up tonight.