dplyr self-join with filter - r

Question

I want to subtract values from a row with label "baseline" from all the otherwise marked items in a long format data frame. It is easy to do this in two steps using a left_join with the "baseline" subset. However, I could not figure out how to combine vas_1 and vas_diff into one chain.
library(dplyr)
# Create test data
n_users = 5
vas = data_frame(
user = rep(letters[1:n_users], each = 3),
group = rep(c("baseline", "early", "late" ),n_users),
vas = round(rgamma(n_users*3, 10,1.4 ))
)
# The above data are given
# Assume some other operations are required
vas_1 = vas %>%
mutate(
vas = vas * 2
)
# I want to put the following into one
# chain with the above
# Use self-join to subtract baseline
vas_diff = vas_1 %>%
filter(group != "baseline") %>%
# Problem is vas_1 here. Using . gives error here
# Adding copy = TRUE does not help
# left_join(. %>% filter(group == "baseline") , by = c("user")) %>%
left_join(vas_1 %>% filter(group == "baseline") , by = c("user")) %>%
mutate(vas = vas.x - vas.y) %>% # compute offset
select(user, group.x, vas) # remove temporary variables
vas_diff

Answer 1

I use anonymous function when . should be used multiple times:
... %>% (function(df) { ... }) %>% ...
Hence, in your case:
vas_diff = vas_1 %>%
filter(group != "baseline") %>%
(function(df) left_join(df, df %>% filter(group == "baseline") , by = c("user"))) %>%
mutate(vas = vas.x - vas.y) %>% # compute offset
select(user, group.x, vas)
(which is not going produce desirable result as describe in comments above, but you it shows how to use anonymous function)
but probably you want this:
vas_diff = vas_1 %>%
left_join(
x = filter(., group != "baseline")
, y = filter(., group == "baseline")
, by = c("user")
) %>%
mutate(vas = vas.x - vas.y) %>% # compute offset
select(user, group.x, vas) # remove temporary variables

Answer 2

Here's a similar option, and a demonstration that you can pass whole pipe chains as arguments to the join. Instead of moving the . inside filter, you can pass . as an argument to eval, then remove unwanted columns in the right-hand side. This is largely just to document this approach for my own purposes.
vas_diff = vas_1 %>%
left_join(x = eval(.) %>%
filter(group != "baseline"),
y = eval(.) %>%
filter(group == "baseline") %>%
select(-group),
by = c("user")) %>%
mutate(vas = vas.x - vas.y) %>% # compute offset
select(user, group, vas)
Does anybody know why you can't simply pass the ., like x = . %>% filter ...... Why do we need to eval(.).