Constructing categorical vector of pure "missing" with prescribed levels

[ablaom]

It would be nice if the following worked, instead of throwing an error:

categorical(fill(missing, 4), levels=["inlier", "outlier"], ordered=true) 

edit corrected as pointed out in next comment

My use case is a machine learning model (an outlier detector) that is semi-supervised. The features are paired with labels indicating outlier/inlier, with possibly some or all labels missing. Even in the case all labels are missing (unsupervised) I still want to extract from the categorical vector the labels to get the user's preference for these labels.

[bkamins]

use vector with proper eltype (as I guess in your code there is a typo and you meant fill(missing, 4) not fill("missing", 4)):

using Missings
categorical(missings(String, 4), levels=["inlier", "outlier"], ordered=true)

[ablaom]

Thanks! I didn't know about missings method.

Still think the request has merit, but feel free to close.

[nalimilan]

It could make sense to take into account the eltype of levels if the eltype of the input is Missing. We could even always call promote_type in both the array and the levels eltypes. Hopefully this only requires adapting a handful of constructors. Feel free to make a PR.