topfeatures per-document option

[adamobeng]

[koheiw]

It is easy to implement this, but how it should look like?

topfeatures(mx, by_document = TRUE)

or

topfeatures(mx, group = docnames(mx))

or even

apply(mx, 1, topfeatures)

[stefan-mueller]

Good idea! I'd prefer the first option as the structure seems to be more similar to the general quanteda syntax.

[kbenoit]

Output would be a sorted list, much like:

sink("/dev/null"); example(textstat_simil); sink()
head(as.list(s1), 2)
# $`1993-Clinton`
# 1997-Clinton   2009-Obama   2013-Obama    2001-Bush    2005-Bush   2017-Trump 
#    0.6660288    0.6288108    0.6278986    0.5333062    0.5092769    0.5033731 
# 
# $`1997-Clinton`
#   2009-Obama   2013-Obama 1993-Clinton    2001-Bush   2017-Trump    2005-Bush 
#    0.6954614    0.6826705    0.6660288    0.6434983    0.6136180    0.5535663 

Possibilities for syntax:

1. For real consistency, we deprecate topfeatures() and provide a textstat_frequency() function.
2. Use the by_document = TRUE as @koheiw suggests
3. Use by = docnames(x) as a default, but allow this to be a grouping variable, including docvar names.

Could of course combine 1 and (2 or 3).

Not pressing on the groups since since we already have a dfm_group(), and the by_document option would give the same result if dfm_group() were run first.