src/piflow_rdf.py
1"""Read RDF triples as entity and relationship streams."""
2
3import pyarrow as pa
4
5TEXT = pa.string()
6ATTRIBUTE = pa.struct([("predicate", TEXT), ("value", TEXT), ("language", TEXT), ("datatype", TEXT)])
7ENTITY = pa.schema([("id", TEXT), ("labels", pa.list_(TEXT)), ("properties", pa.list_(ATTRIBUTE))])
8RELATION = pa.schema([("source", TEXT), ("predicate", TEXT), ("target", TEXT)])
9
10
11def parse_rdf(data, syntax):
12 from rdflib import BNode, Graph, Literal, RDF
13
14 if syntax not in {"nt", "turtle"}:
15 raise ValueError("syntax must be nt or turtle")
16 graph = Graph()
17 graph.parse(data=data.decode("utf-8"), format=syntax, publicID="urn:piflow:input")
18 if len(graph) > 100_000:
19 raise ValueError("RDF graph exceeds 100,000 triples")
20
21 def resource(term):
22 return "_:" + str(term) if isinstance(term, BNode) else str(term)
23
24 entities = {}
25 relations = []
26 for a, predicate, b in sorted(graph, key=lambda row: [part.n3() for part in row]):
27 origin = resource(a)
28 entity = entities.setdefault(origin, dict(id=origin, labels=[], properties=[]))
29 if isinstance(b, Literal):
30 entity["properties"].append(dict(predicate=str(predicate), value=str(b),
31 language=b.language, datatype=str(b.datatype) if b.datatype else None))
32 elif predicate == RDF.type:
33 entity["labels"].append(resource(b))
34 else:
35 target = resource(b)
36 entities.setdefault(target, dict(id=target, labels=[], properties=[]))
37 relations.append(dict(source=origin, predicate=str(predicate), target=target))
38 return {"entityOut": pa.Table.from_pylist([entities[name] for name in sorted(entities)], schema=ENTITY),
39 "relationshipOut": pa.Table.from_pylist(relations, schema=RELATION)}