tidyverse hints

tidy_hint.R

?dplyr::context
# gives us
cur_column() - current column name!
cur_<press tab>
n()

disp <- 10
mtcars %>% mutate(disp = .data$disp * .env$disp)
mtcars %>% mutate(disp = disp * disp)
?.data

?c_across

across(...., .names='{.col}_{.fn}')

# a pair of columns problem
# 1. there are pairs of columns like transaction_type _duration and _n
# we need to pick some ot the pairs, not all of them
# then divide every corresponding _dur to _n column
# solution 1
reduce(striped col names, .init=df, .f=
       ~mutate(.x, '{.y}_dur_div_n' = get(glue('{.y}_dur')) / get(glue('{.y}_n')) )
# solution 2
mutate(across.. cur_column()

#   name     mae_nv     n mae_gain
#  <chr>     <dbl> <dbl>    <dbl>
# 1 adv        23.8  1123    0.036
# 2 odds       23.8  1123    0.077
# 3 adv_21+    22.6   297   -0.022
# 4 odds_21+   22.6   297    0.028
# group_by n, subtract mae_gain in each group
group_by(n) %>% 
    group_map(~deframe(mutate(., name=str_remove(name, '_.+'))) %>% 
                {.['adv'] - .['odds']})
                
# 2. head(x, -2) is a very useful feature allowing to select all but 2 last 
# values. It works even for whole tibbles. 
# At the time of writing, unfortunately, there is no such alternative for grouped tibble
# applying it separately to each group. My solution
tibble(gr=c(1,1,1,2,2,2,2), val=1:7) %>% 
      group_by(gr) %>% 
      filter(n() - row_number() + 1 > 2) %>% 
      ungroup()
      
# 3. fast time cv can be done with nesting and lags

# 4. group_by()/rowwise() %>% group_map() is bad
# Especially when you need a grouping value in the result
# Replace it with nesting
df %>% 
  nest(data=-smth) %>% 
  mutate(res=map(data, your_func), .keep='unused')
       
# 5. to inject a vectorized lambda function in a long pipeline 
1:4 %>% exec(as_mapper(~. + 1), .)

# 6. chr to tbl with delay, by batches
x %>% 
       enframe() %>% 
       group_by(batch=ceiling(name/10)) %>% 
       summarise(ids=list(value)) %>% 
       pluck('ids') %>% 
       imap_dfr(~...