Hello there,
my dataset contains columns that also include multi-valued attributes like in this example column:
"EC_Aaa123"
"EC_Xxx, EC_Yyyy, EC_Z"
"EC_Bbbb567"
(* You can think of proteins that can be labeled with one or multiple EC numbers that come from a hierarchical tree-like taxonomy. I am actually using some different taxonomies, but the principle is the same)
I know that ARX cannot generalize such multi-value strings, so I did it already in a preprocessing step. That is, I used preprocessing to simplify all multi-valued strings as much as possible with respect to the underlying taxonomy, using a most recent common ancestor approach.
Now I want to provide ARX this column along with a mock-up taxonomy, like so:
EC_Aaa123; EC_Aaa; EC_A; ****
EC_Bbb567; EC_Bbb; EC_B; ****
EC_Xxx,EC_Yyyy,EC_Z; ****; ****; ****
So, the idea is to give ARX the freedom to generalize at least the single-value attributes as best as possible, or to suppress the multi-valued ones. By the way, the items in multi-valued strings appear in sorted order, to avoid ambiguity.
This approach works except for some columns. ARX complains about missing values, but I don't know why. I checked the taxonomies several times.
Are there any latent restrictions that apply? For example, on the length of the strings in the input columns?
I really hope you can help me, as I already invested two weeks into the intricate preprocessing steps of the dataset. Unfortunately, I am not allowed to share that data.