`group_by()` doesn't work with `atlas_counts (type = "species")`

[mjwestgate]

This works:

galah_call() |>
    filter(year == 2025) |>
    atlas_counts(type = "species")
# A tibble: 1 × 1
  count
  <int>
1 35292

But this doesn't:

galah_call() |>
    filter(year == 2025) |>
    group_by(cl22) |>
    atlas_counts(type = "species")
Error in `dplyr::bind_cols()` at galah-R/R/collect_species_count.R:9:5:
! Can't recycle `..1` (size 11) to match `..2` (size 3).
Run `rlang::last_trace()` to see where the error occurred.

This could be viewed as a simple bug, but there might be a deeper problem here in that recent versions use group_by(speciesID) as equivalent to atlas_species(). This leads to a problem where we are effectively using group_by() twice, to do different things:

• once to tell us the number of categories present (species)
• another to give the levels of categories (states and territories)

Given that these are distinct operations, this suggests that group_by() isn't solely the solution, and something more dplyr-like might be needed to distinguish them. Not sure what that looks like rn but might involve summarise.data_request() or similar.

[mjwestgate]

@daxkellie points out that there is a solution here using summarise(), though arguably not a pretty one:

result <- galah_call() |>
  identify("perameles") |>
  filter(year %in% c(2004, 2005)) |>
  select(group = "basic", year, species) |>
  atlas_occurrences()
#> Request for 388 occurrences placed in queue
#> Current queue length: 1
#> ----
#> Downloading

# one summarise
result |>
  group_by(year, species) |>
  summarise(count = n())
#> `summarise()` has grouped output by 'year'. You can override using the
#> `.groups` argument.
#> # A tibble: 6 × 3
#> # Groups:   year [2]
#>    year species              count
#>   <dbl> <chr>                <int>
#> 1  2004 Perameles gunnii        12
#> 2  2004 Perameles nasuta       199
#> 3  2004 Perameles pallescens     7
#> 4  2005 Perameles gunnii         1
#> 5  2005 Perameles nasuta       168
#> 6  2005 Perameles pallescens     1

# two summarise
result |>
  group_by(year, species) |>
  summarise(count = n()) |>
  summarise(count = n())
#> `summarise()` has grouped output by 'year'. You can override using the
#> `.groups` argument.
#> # A tibble: 2 × 2
#>    year count
#>   <dbl> <int>
#> 1  2004     3
#> 2  2005     3

i.e. one summarise() equivalent to count()/atlas_counts() and two summarise() giving expected behaviour for atlas_counts(type = "species")

One possible advantage here is that it also provides logical syntax for getting both numbers of species and occurrences:

result |>
+     group_by(year, species) |>
+     summarise(count = n()) |>
+     summarise(n_species = n(),
+               n_records = sum(count))
`summarise()` has grouped output by 'year'. You can override using the `.groups` argument.
# A tibble: 2 × 3
   year n_species n_records
  <dbl>     <int>     <int>
1  2004         3       218
2  2005         3       170

[mjwestgate]

Just highlighting this section from ?dplyr::count()

df %>% count(a, b) is roughly equivalent to df %>% group_by(a, b) %>% summarise(n = n())

i.e. for consistency with dplyr, it would be ideal not just to support summarise.data_request(), but also to accept a ... argument in count.data_request() as a substitute for using group_by() |> atlas_counts(). It is unclear whether this would constitute a breaking change by altering interaction between group_by.data_request and count.data_request.

[mjwestgate]

A different approach here might be to see what syntax in dplyr would convert occurrences into the formats currently supplied by atlas_species(), atlas_counts() etc. Noting that the syntax question has already been raised in issue #275.

# load libraries
library(galah)
library(dplyr)

# get example occurrence dataset
galah_config(email = "an-email@@email.com")
testdata <- galah_call() |>
  identify("perameles") |>
  select(speciesID, group = "basic") |>
  filter(year == 2020) |>
  collect()

# number of records
testdata |> 
  count()

# number of records per species
testdata |>
  count(speciesID) # noting this is currently supported via `group_by() |> count()`, which is also valid dplyr syntax

# number of species (aka `atlas_counts(type = "species")`)
testdata |>
  summarize(count = n_distinct(speciesID))

# one row per species (aka `atlas_species()`)
testdata |>
  distinct(speciesID, .keep_all = TRUE)
# Noting that this requires care with `select()` to return sensible results

# show only which species are present in the query (aka `show_values()`)
testdata |>
  distinct(speciesID, .keep_all = FALSE)

# get species (and optionally, record) counts grouped by a second factor
testdata |>
  group_by(basisOfRecord) |>
  summarize(n_records = n(),
            n_spp = n_distinct(speciesID))

# get one row per species, showing number of records
testdata |>
  group_by(speciesID) |>
  mutate(count = n()) |>
  distinct(.keep_all = TRUE)

Commentary

• summarise() and mutate() are very flexible and are used here in a more restricted sense; probably only possible to support a subset of behaviours
• distinct() is useful because it supports one row per column and retains non-selected columns, but is otherwise an inexact match to atlas_species(). We'd have to actively prevent requesting non-supported fields (via select()) and limit to a single variable.

[mjwestgate]

An easier solution than the above is to implement only add_count() and distinct() from dplyr, as follows:

# number of species (total)
# aka atlas_counts(type = "species")
galah_call() |>
  distinct(taxonConceptID) |>
  count()

# one row per basisOfRecord, showing number of species
# aka atlas_counts(type = "species") with group_by()
galah_call()  |>
  group_by(basisOfRecord) |>
  distinct(taxonConceptID) |>
  count()

# atlas_species() with added record counts
galah_call() |>
  group_by(taxonConceptID) |>
  add_count() |> 
  distinct(taxonConceptID, .keep_all = TRUE)