dplyr group_by and cummean functions

Question

I expected the code below to output a data frame with three rows, each row representing the cumulative mean value of mpg after calculating the mean for each group of cyl:

library(dplyr)
mtcars %>%
arrange(cyl) %>%
group_by(cyl) %>%
summarise(running.mean.mpg = cummean(mpg))

This is what I expected to happen:

mean_cyl_4 <- mtcars %>% 
filter(cyl == 4) %>%
summarise(mean(mpg))

mean_cyl_4_6 <- mtcars %>% 
filter(cyl == 4 | cyl == 6) %>%
summarise(mean(mpg))

mean_cyl_4_6_8 <- mtcars %>% 
filter(cyl == 4 | cyl == 6 | cyl == 8) %>%
summarise(mean(mpg))

data.frame(cyl = c(4,6,8), running.mean.mpg = c(mean_cyl_4[1,1], mean_cyl_4_6[1,1], mean_cyl_4_6_8[1,1]))

  cyl running.mean.mpg
1   4     26.66364
2   6     23.97222
3   8     20.09062

How come dplyr seems to ignore group_by(cyl)?

Answer 1

require("dplyr")

mtcars %>%
  arrange(cyl) %>%
  group_by(cyl) %>%
  mutate(running.mean.mpg = cummean(mpg)) %>%
  select(cyl, running.mean.mpg)

# Source: local data frame [32 x 2]
# Groups: cyl
# 
# # cyl running.mean.mpg
# # 1    4         22.80000
# # 2    4         23.60000
# # 3    4         23.33333
# # 4    4         25.60000
# # 5    4         26.56000
# # 6    4         27.78333
# # 7    4         26.88571
# # 8    4         26.93750

For the sake of experimentation, this would also work with data.table. I mean, you have to load dplyr also to have cummean() available.

require("data.table")
DT <- as.data.table(mtcars)
DT[,j=list(
  running.mean.mpg = cummean(mpg)
  ), by="cyl"]