Use rle to group by runs when using dplyr

Question

In R, I want to summarize my data after grouping it based on the runs of a variable x (aka each group of the data corresponds to a subset of the data where consecutive x values are the same). For instance, consider the following data frame, where I want to compute the average y value within each run of x:

(dat <- data.frame(x=c(1, 1, 1, 2, 2, 1, 2), y=1:7))
#   x y
# 1 1 1
# 2 1 2
# 3 1 3
# 4 2 4
# 5 2 5
# 6 1 6
# 7 2 7

In this example, the x variable has runs of length 3, then 2, then 1, and finally 1, taking values 1, 2, 1, and 2 in those four runs. The corresponding means of y in those groups are 2, 4.5, 6, and 7.

It is easy to carry out this grouped operation in base R using tapply, passing dat$y as the data, using rle to compute the run number from dat$x, and passing the desired summary function:

tapply(dat$y, with(rle(dat$x), rep(seq_along(lengths), lengths)), mean)
#   1   2   3   4 
# 2.0 4.5 6.0 7.0 

I figured I would be able to pretty directly carry over this logic to dplyr, but my attempts so far have all ended in errors:

library(dplyr)
# First attempt
dat %>%
  group_by(with(rle(x), rep(seq_along(lengths), lengths))) %>%
  summarize(mean(y))
# Error: cannot coerce type 'closure' to vector of type 'integer'

# Attempt 2 -- maybe "with" is the problem?
dat %>%
  group_by(rep(seq_along(rle(x)$lengths), rle(x)$lengths)) %>%
  summarize(mean(y))
# Error: invalid subscript type 'closure'

For completeness, I could reimplement the rle run id myself using cumsum, head, and tail to get around this, but it makes the grouping code tougher to read and involves a bit of reinventing the wheel:

dat %>%
  group_by(run=cumsum(c(1, head(x, -1) != tail(x, -1)))) %>%
  summarize(mean(y))
#     run mean(y)
#   (dbl)   (dbl)
# 1     1     2.0
# 2     2     4.5
# 3     3     6.0
# 4     4     7.0

What is causing my rle-based grouping code to fail in dplyr, and is there any solution that enables me to keep using rle when grouping by run id?

Answer 1

One option seems to be the use of {} as in:

dat %>%
    group_by(yy = {yy = rle(x); rep(seq_along(yy$lengths), yy$lengths)}) %>%
    summarize(mean(y))
#Source: local data frame [4 x 2]
#
#     yy mean(y)
#  (int)   (dbl)
#1     1     2.0
#2     2     4.5
#3     3     6.0
#4     4     7.0

It would be nice if future dplyr versions also had an equivalent of data.table's rleid function.

---

I noticed that this problem occurs when using a data.frame or tbl_df input but not, when using a tbl_dt or data.table input:

dat %>% 
    tbl_df %>% 
    group_by(yy = with(rle(x), rep(seq_along(lengths), lengths))) %>%
    summarize(mean(y))
Error: cannot coerce type 'closure' to vector of type 'integer'

dat %>% 
    tbl_dt %>% 
    group_by(yy = with(rle(x), rep(seq_along(lengths), lengths))) %>%
    summarize(mean(y))
Source: local data table [4 x 2]

     yy mean(y)
  (int)   (dbl)
1     1     2.0
2     2     4.5
3     3     6.0
4     4     7.0

I reported this as an issue on dplyr's github page.

Answer 2

If you explicitly create a grouping variable g it more or less works:

> dat %>% transform(g=with(rle(dat$x),{ rep(seq_along(lengths), lengths)}))%>%                                   
 group_by(g) %>% summarize(mean(y))
Source: local data frame [4 x 2]

      g mean(y)
  (int)   (dbl)
1     1     2.0
2     2     4.5
3     3     6.0
4     4     7.0

I used transform here because mutate throws an error.