Document TDL inverse property feature extraction performance improvements

[Copilot]

@Demirrr requested opening an issue to track performance improvements for TDL's inverse property feature extraction method, which has O(P × I) complexity causing scalability issues on large knowledge graphs.

Context

The _extract_inverse_property_features method (tree_learner.py:374-413) iterates through all object properties and all individuals for each individual being processed:

for obj_prop in self.knowledge_base.get_object_properties():  # P iterations
    for other_ind in self.knowledge_base.individuals():  # I iterations
        # Check if individual is object of property assertion

For a KB with 1,000 individuals and 50 properties: 50,000 iterations per individual processed.

Changes

• Created comprehensive issue documentation in /tmp/performance-improvement-issue.md
• Documented optimization strategies:
  • Direct KB queries for assertions by object
  • Class-level caching of property assertions
  • Sparse assertion indexing at initialization
  • Lazy feature evaluation
• Replied to code review comment explaining access limitations for creating GitHub issues directly

Note

Cannot create GitHub issues programmatically from this environment. Issue template provided for manual creation with all technical details, proposed solutions, and implementation considerations.

---

💬 We'd love your input! Share your thoughts on Copilot coding agent in our 2 minute survey.