Closed HenrikBengtsson closed 3 years ago
Thanks @HenrikBengtsson for the heads-up.
Because we never compute rowRanks on any vector that could have a tie, I think we can keep code as is (so going with the new default).
As in SAMseq, we add runif
to anything we compute ties on.
Ok, good. However, I'll be deprecating, and then defunct:ing, the case when missing(ties.method)
is true, for quite a while, before changing the default. This is to alert as many people as possible, so they can update any scripts using it. I wanna be conservative and minimize the risk for the change to go unnoticed "out there". So, you'll eventually have to specify the argument ... at least for quite a while.
Oh I see, I'll switch to ties.method = "average"
then in devel branch.
matrixStats will align the default
ties.method
forcolRanks()
androwRanks()
with that ofbase::rank()
, cf. https://github.com/HenrikBengtsson/matrixStats/issues/142. Because of this, I ran a few reverse dependency checks to see what could break, and I spotted fishpond. It doesn't specifyties.method
, so its results will be affected when we change the default:https://github.com/mikelove/fishpond/search?q=rowRanks