Alan-Jowett / LexonGraph / 27076521757

81%

Build Type

push

github

Committed by web-flow

Commit Message

Rework directional PCA for streaming clustering (#61)

* Rework directional PCA for streaming clustering

Update the directional PCA spec package and crate to use the shared streaming clustering contract instead of the retired block-store layer API.

The implementation now provides a native streaming trainer/classifier, enforces exact-K failure behavior, and removes dead block-store-oriented code and tests.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* Address PR review feedback

Replace full baseline pass retention with a deterministic SHA-256 pass fingerprint and remove an unnecessary dead_code allowance in test support.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* Trim directional PCA pass allocations

Avoid cloning the trained model when storing pass state and compute group centroids directly from indexed embeddings instead of cloning temporary vectors.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* Align directional PCA config semantics

Allow non-negative variance exponents above 1.0 to match the approved spec and use fixed-width u64 encoding in pass fingerprints for portable deterministic replay checks.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* Clarify directional PCA gamma authority

Make the crate spec authoritative for parameter-domain rules and state explicitly that this streaming crate accepts finite, non-negative gamma values.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

---------

Co-authored-by: Alan Jowett <alan.jowett@microsoft.com>
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Coverage Stats

314 of 397 new or added lines in 1 file covered. (79.09%)

9 existing lines in 2 files now uncovered.

7581 of 8973 relevant lines covered (84.49%)

217.38 hits per line