epi_slide* no column behavior

[dsweber2]

When I don't specify a column, epi_slide and epi_slide_mean have pretty different behavior. Example command epi_slide_mean(before = 6L) or epi_slide(mean, before = 6L).

epi_slide_mean outputs a Can't recycle input of size 0 to size 1 error, while epi_slide just returns a ton of NA values.

Ideally they would either

1. assume the correct target (in the case of 1 non-key column there's an obvious choice)
2. throw an error saying we need to specify a target (probably the right behavior if there's 2+ columns)

edit: second epi_slide_mean should've been just epi_slide

[brookslogan]

Couple notes:

• The cause of the epi_slide_mean() behavior is result_col_names <- paste0("slide_value_", col_names_chr) which doesn't properly vectorize to a 0-length result in the case of a 0-length col_names_chr. [without requiring R >= 4.0.1 and specifying recycle0 = TRUE]
• In the epi_slide() example, (i) we issue a warning that mean doesn't look like a proper function to be sliding, (ii) mean generates a bunch of NAs + warnings from calculating the mean across an epi_df chunk + group key + ref time value, including non-numeric columns of the epi_df chunk or group key, and (iii) default R behavior collapses all these warnings into a less-noticeable (especially with my editor's coloring settings) "There were 50 or more warnings (use warnings() to see the first 50)".

We can add additional validation/defaults for the epi_slide_mean() case. To stop the epi_slide() case, we'd probably need to turn that warning into an error.

[dsweber2]

Ah, I guess I was a bit confused about epi_slide working after including the column name. The lack of parallel is unfortunate; it would be nice if epi_slide accepted this sort of arg that switched it to operating on the specific column.

[brookslogan]

The tidyverse way would be to use across(). But that may be in the group of things that really delve into dplyr internals and aren't compatible with (simple usage of) the rlang analogues (that might not even back dplyr ops despite sounding that way??), so allowing across() in epi_slide() might not be simple.

[brookslogan]

In #521 I'm just going for

throw an error saying we need to specify a target

in all cases, plus hinting them to use epi_slide_mean etc. We could reconsider whether we want to do some sort of special behavior when there's only 1 sensible value column, or maybe doing something for every value column if they are all sensible. But I'm just trying to make sure we have a hard & helpful error as a starting point.

[brookslogan]

#521 lacks the epi_slide_mean fix [0 cols being fed into epi_slide_mean. The epi_slide part is "addressed" with an error message.]

[brookslogan]

Remaining TODO is the epi_slide_{opt,mean,sum} adjustment. We have some similar logic in revision_summary() we could mirror / borrow code from.

#469 (comment) & maybe elsewhere we have a alternative proposal for the default operating column(s).