Change output of pariwise functions when nothing to compare, to allow for summarize use

[wfulp]

Currently piarwise functions return NULL when only one group to compare. This breaks if using in group_by --> summarize. Could we change to data.frame of NAs instead?

[master] > control_testing_data %>%
+   group_by(day_plot, measurement, sample_type) %>%
+   summarise(pairwise_test_cont(x = magnitude, group = study_grouped),
+             .groups = 'drop') 
Error: Problem with `summarise()` input `..1`.
x Input `..1` must be a vector, not NULL.
i Input `..1` is `pairwise_test_cont(x = magnitude, group = study_grouped)`.
i The error occurred in group 1: day_plot = "Day 1", measurement = "sgmRNA", sample_type = "bal".
Run `rlang::last_error()` to see where the error occurred.

[master] > tmp_dat <- control_testing_data %>% 
+   filter(day_plot == "Day 1", measurement == "sgmRNA", sample_type == "bal")

[master] > pairwise_test_cont(x = tmp_dat $magnitude, group = tmp_dat $study_grouped)
NULL

[mayerbry]

Can you provide a reprex for this issue?

I have two concerns:

1. how experimental is this new stacking feature with summarize? What is it actually doing on the backend (i.e., how do you interpret "Input ..1 must be a vector, not NULL.")?
2. we don't want to break old workflows, I'm guessing we return NULLs because of list stacking. Maybe not the best way to do things, but a different conversation :). The NA data.frame might be fine, but would be nice to experiment with a reprex.

[wfulp]

Here is the wrapper I used as a workaround:

aa = function(...) {
  tmp_results = pairwise_test_cont(...)
  if (is.null(tmp_results)) {
    tmp_results = data.frame('Comparison' = NA, 
                             "SampleSizes" = NA,       
                             "Median_Min_Max" = NA,     
                             "Mean_SD" = NA,            
                             "MagnitudeTest" = NA,  
                             "PerfectSeperation" = NA)
  }
  tmp_results
}

[mayerbry]

some tidyr workarounds (Jimmy confirmed worked):

control_testing_data %>%
  group_by(day_plot, measurement, sample_type) %>%
  summarise(x = list(pairwise_test_cont(x = pmax(magnitude, 86), group = study_grouped)),
               .groups = 'drop') %>%
  unnest(cols = x)

more explicit

control_testing_data %>%
  group_by(day_plot, measurement, sample_type) %>%
  summarise(x = list(pairwise_test_cont(x = pmax(magnitude, 86), group = study_grouped)),
               .groups = 'drop') %>%
  filter(!is.null(x)) %>%
  unnest(cols = x)

[wfulp]

If a change is made a requirement is it being backwards compatible, specifically that it runs the same with nest, do, and group_modify functions