--individuals exclude does not exclude all individuals

[pfabry]

Using the --individuals exclude parameter with the extract function does not remove all individuals.

See example here: https://github.com/pfabry/ROBOT_individuals_exclude

I used obi.owl as test (https://github.com/obi-ontology/obi) and tried the following commands:

 robot extract --input obi.owl --method TOP --term BFO:0000001 --output obiV1.owl
 robot extract --input obi.owl --method TOP --term BFO:0000001 --individuals exclude --output obiV2.owl

In the first case obiV1.owl has 282 individuals
In the second case, obiV2.owl has 44 individuals
I found the same behavior with other ontologies (OMO, OMRSE for example).
The remaining individuals have different namespaces (OBI, UO, IAO, etc.). However, they didn't have a class type associated anymore. For a given individual present in both obiV1 and obiV2, it will have a class type in obiV1 but not anymore in obiV2.

[jamesaoverton]

Are the remaining individuals used in the logical axioms of classes, e.g. IAO:0000078 curation status specification? I suspect so, in which case there's no way to get rid of them without removing those class axioms.

We could update the docs to clarify this.

[pfabry]

Indeed they are. As a workaround I preprocess the input file with remove --select individuals with the "inconvenient" of having to create custom import module in ODK.
On a side note, I was expecting that remove --axioms abox and remove --select individuals both produce the same result but it seems that --axioms abox doesn't select any abox related axioms in the input ontology.

[matentzn]

Maybe can add a new exclusion code to extract --individuals exclude-incl-assertions - this is natural for many extraction use cases. A more complex solution would be a parameter --ignore individuals which explicitly does not remove them, but does the following:

1. Remove all axioms which contain 1 or more individuals from O (O1)
2. Extract module M1 from O1.
3. Extract filter module M2 from O (!, not O1).
4. Merge M1 and M2.

The advantage is with this latter solution is that the individual axioms do not impact the SLME module extraction, but are still included in the output if the user explicitly says so.

Neither of these would break the existing functionality.

[pfabry]

Maybe can add a new exclusion code to extract --individuals exclude-incl-assertions - this is natural for many extraction use cases.

Shouldn't we get the same result with remove --select individuals or remove --axioms abox ?

[matentzn]

Shouldn't we get the same result with remove --select individuals or remove --axioms abox ?

Not exactly:

EquivalentClass(:A,{:i, :j}), where :i and :j are individuals, is not an Abox axiom ({:i, :j} is a nominal expression that basically defines a set in terms of some specific individuals.