linkml-datalog
Validation and inference over LinkML instance data using souffle
Requirements
This project requires souffle
After installing souffle, install the python here is a normal way.
Until this is released to pypi:
poetry install
Running
Pass in a schema and a data file
poetry run python -m linkml_datalog.engines.datalog_engine -d tmp -s personinfo.yaml example_personinfo_data.yaml
The output will be a ValidationReport object, in yaml
e.g.
- type: sh:MaxValue
subject: https://example.org/P/003
instantiates: Person
predicate: age_in_years
object_str: '100001'
info: Maximum is 999
Currently, to look at inferred edges, consult the directory you specified in -d
E.g.
tmp/Person_grandfather_of.csv
Will have a subject and object tuple P:005 to P:001
How it works
- Schema is compiled to Souffle DL problem (see generated schema.dl file)
- Any embedded logic program in the schema is also added
- Data is converted to generic triple-like tuples (see
*.facts) - Souffle executed
- Inferred validation results turned into objects
Assuming input like this:
classes:
Person:
attributes:
age:
range: integer
maximum_value: 999
The generated souffle program will look like this:
.decl Person_age_in_years_asserted(i: identifier, v: value)
.decl Person_age_in_years(i: identifier, v: value)
.output Person_age_in_years
.output Person_age_in_years_asserted
Person_age_in_years(i, v) :-
Person_age_in_years_asserted(i, v).
Person_age_in_years_asserted(i, v) :-
Person(i),
triple(i, "https://w3id.org/linkml/examples/personinfo/age_in_years", v).
validation_result(
"sh:MaxValueTODO",
i,
"Person",
"age_in_years",
v,
"Maximum is 999") :-
Person(i),
Person_age_in_years(i, v),
literal_number(v,num),
num > 999.
Motivation / Future Extensions
The above example shows functionality that could easily be achieved by other means:
- jsonschema
- shape languages: shex/shacl
In fact the core linkml library already has wrappers for these. See working with data in linkml guide.
However, jsonschema in particular offers very limited expressivity. There are many more opportunities for expressivity with linkml.
In particular, LinkML 1.2 introduces autoclassification rules, conditional logic, and complex expressions -- THESE ARE NOT TRANSLATED YET, but they will be in future.
For now, you can also include your own rules in the header of your schema as an annotation, e.g the following translates a 'reified' association modeling of relationships to direct slot assignments, and includes transitive inferences etc
has_familial_relationship_to(i, p, j) :-
Person_has_familial_relationships(i, r),
FamilialRelationship_related_to(r, j),
FamilialRelationship_type(r, p).
Person_parent_of(i, j) :-
has_familial_relationship_to(i, "https://example.org/FamilialRelations#02", j).
Person_ancestor_of(i, j) :-
Person_parent_of(i, z),
Person_ancestor_of(z, j).
Person_ancestor_of(i, j) :-
Person_parent_of(i, j).
See tests for more details.
In future these will be compilable from higher level predicates
Background
See #196
1 Jan 19, 2022
359 Dec 22, 2022
1 Jan 16, 2022
27 Nov 1, 2022
6 Sep 7, 2022
791 Jan 4, 2023
40 Dec 12, 2022
184 Dec 27, 2022
3 Oct 3, 2022
1 Feb 18, 2022
2 Nov 16, 2021
2 Feb 16, 2022
4 Mar 06, 2022
0 Dec 04, 2021
3.8k Jan 05, 2023
971 Dec 25, 2022
3.6k Jan 09, 2023
95 Dec 13, 2022
69 Dec 29, 2022
2k Dec 29, 2022
447 Dec 25, 2022
102 Nov 10, 2022
24 Dec 14, 2022
1 Jan 26, 2022
32 Apr 25, 2022
126 Dec 28, 2022
863 Jan 08, 2023
8k Dec 29, 2022