# PlackettLuce TODO

## general

* PlackettLuce line 351 - is it necessary to go through set sizes of 1? Maybe 2:nc okay here
* is it necessary to compute X when you have starting values already?
* `A[item] <- unname` where are the names comming from
 - can I do reorder = FALSE (and not sort item's earlier?)
* split up vignette now arxiv article published
 - basic introduction
 - comparison with other packages
 - trees
 - adherence/MAP estimation
* check it works when missing observed ties of given order, e.g. observe ties order 3 but none of order 2

We have not investigated this question in depth, because we
have only recently come across data sets with ties of order greater than 2 in
practice. Often these data sets are derived from ratings, which are not
currently supported by PlackettLuce in general. This is because a rating scale
of r values implies free choices only for the first (r - 1) ranks, while
PlackettLuce assumes all ranks are free choices unless there is only 1 item.
Supporting ratings data is a

We have not investigated the practical support for higher order ties in depth,
since most partial rankings data we have come across in practice has ties of
low order. Ties of higher order can occur in rankings derived from ratings,
however this is a topic of future work, since it requires special treatment
(items are tied bottom due to the limited rating scale, not by free choice).
