Skip to content

Latest commit

 

History

History
61 lines (45 loc) · 2.42 KB

TODO.md

File metadata and controls

61 lines (45 loc) · 2.42 KB
Raw

TODO

Model

  • Statement patterns: We could start introducing patterns via a statement pattern, i.e., a statement template together with a variable constraint.

  • Add AnnotatedStatement object to carry statement together with its set of annotation records.

  • BUG: Fix the internal representation of dates and times. Maybe we shouldn't use Python's datatime.

  • Filter (new feature): Add support for "negation". We can compile the negation of an atomic v, i.e., ~v, as FILTER(?x != v). And we can compile the negation of a snak S, i.e., ~S, as a FILTER NOT EXISTS.

  • Fingerprint (normalization): We can use the distributive laws to decompose complex fingerprint. E.g., 𝜂[A∧(B∨(C∧(D∨E)))] ⤳ 𝜂[A∧B]∨𝜂[A∧(C∧(D∨E))] ⤳ (A∧B)∨𝜂[(X≔A∧C)∧(D∨E)] ⤳ (A∧B)∨𝜂[X∧(D∨E)] ⤳ (A∧B)∨𝜂[X∧D]∨𝜂[X∧E] ⤳ (A∧B)∨(A∧C∧D)∨(A∧C∧E). Now we know that the value mask of the original formula is equal to the mask of (A∧B) | (A∧C∧D) | (A∧C∧E); also we can break the original query into three queries (executed in parallel). We should restrict this type of normalization to non-VALUES clauses. One possible way to do this is to introduce a new kind of fingerprint ValuesFingerprint/OneOfFingepprint which behaves as | but aggregates only or-ed value fingerprints.

Compiler

  • Filter compiler (optimization): Aggregate snaks with the same property.

  • Use subqueries to implement fingerprints.

Codec

  • Wikidata RDF: Write a new Wikidata RDF encoder for model classes. The proposed AnnotatedStatement object (see above) can help us here.

  • Repr: Replace eval() by a proper parser (via lark).

Store

  • Add async API for Store. See the technique used in kif_lib.vocabulary.wd.downloader.

  • Mapper: We could use statement templates as the unity of mappings. That is, an entry in the mapping could be a pair (p,push) where p is a statement pattern and push() is a function that takes a query builder as argument and pushes the SPARQL fragment that matches p in the target graph. The push() function should also update the substitution with variables that occur in p, so that the final substitution will automatically translate from the target graph to KIF model objects.

  • RDF: Allow user to construct a store by passing statement objects (requires Wikidata RDF encoder).