Closed nickrobinson251 closed 3 years ago
I'd be okay with retaining the missing
in this case. One thing I've been considering is that if we split out an Iterators interface then we'd likely want to use an OnlineStat
instead. Unfortunately, this would just use that default value (e.g., value(Mean()) -> 0.0
) which also might not be what we want? I guess in that case we could check nobs(value)
before returning.
Alright, #69 fixed this by having impute(data::AbstractArray{Missing}, imp)
just return in the input data
cause most imputation methods are going to need at least some non-missing values to work properly.
Here are some common functions we may want to use with
fill
but which (arguably) have awkward behaviour if all data ismissing
-- stemming fromfill
usingdrop
and subsequently ending up calling the function on a empty array:(Impute v0.3, tested on Julia v1.2)
Apologies for not having a solution here... but some question we might want to think about
missings
, a lalocf
if the first element ismissing
?