I have a dataframe samples which has 50 rows, 10 for each of 5 samples. In this df, the sample no is labelled by a column sample_no.

I also have a function f which takes a sample and returns a one-row summary data frame. When I write:

samples %>%
group_by(sample_no) %>%
dplyr::summarise( d = f(.) ) %>%
unpack(cols = d)```

f seems to see all samples at once (50 rows) rather than one sample at a time (10 rows). By contrast

f(samples %>% filter(sample_no == 1))

Sees 10 rows as I would expect.

Is it the case that only certain special functions can be used inside summarise, and that those functions have to be ‘group-aware’? Or is something else happening?

MWE:

library(tidyverse)

SAMPLE_SIZE <- 10
NSAMPLES <- 5

fresh_sample <- function() {
  data.frame(F = rnorm(SAMPLE_SIZE, 0, 1))
}

# I'd also love to know if there is a neater way to do this...
samples <- do.call(rbind, lapply(1:NSAMPLES, function(k) {
  fresh_sample() %>% mutate(sample_no = k)
}))

# I don't actually need to compute dimensions. 
# The actual computation I want to apply to a sample is much more complex.
dims <- function(sample) {
  data.frame(nrow = nrow(sample), ncol = ncol(sample))
}

# This returns a 1-row df, with      nrow = 10, ncol = 2
dims(samples %>% filter(sample_no == 1))

# This returns a 5-row df, each with nrow = 50, ncol = 2
samples %>%
  group_by(sample_no) %>%
  dplyr::summarise(d = dims(.)) %>%
  unpack(cols = d)