Skip to content

TODO.md

Latest commit

 

History

History
27 lines (18 loc) · 1.18 KB

TODO.md

File metadata and controls

27 lines (18 loc) · 1.18 KB

TODO — Known Bugs

Medium

#7 Missing m_ZeroR guard in distributionForAttributes (FasterForest and FasterForest2) — WONTFIX

Files:

  • src/main/java/cz/siret/prank/fforest/FasterForest.java:719
  • src/main/java/cz/siret/prank/fforest2/FasterForest2.java:772

When the training data has only a class attribute (no features), buildClassifier creates a m_ZeroR fallback model and returns early without initializing m_bagger. The distributionForInstance() method correctly checks if (m_ZeroR != null) and delegates to the fallback. However, distributionForAttributes() does not check — it directly calls m_bagger.distributionForAttributes(...), which throws a NullPointerException because m_bagger is null.

This also affects all code paths that go through distributionForAttributes:

  • predict() in both FasterForest (line 829) and FasterForest2 (line 964)
  • predictForBatch() in both, which calls getTrees() → m_bagger.getClassifiersAsTrees()
  • toFlatBinaryForest() in both, which also accesses m_bagger

Not fixing — the m_ZeroR fallback and Weka dependency are planned for removal.

See RESOLVED.md for fixed and documented items.