v3.0.0: N² → N-linear memory optimizations + GPU backend + vendor SAM#174

Merged

atarashansky merged 19 commits intomainfrom

Mar 12, 2026

Commits on Mar 12, 2026

fix: pre-existing bugs and dead code (Phase 0.2)
atarashansky
committed
feat(sam): vendor SAM algorithm as src/samap/sam/ (Phase 0.3)
atarashansky
committed
test: golden regression test for 3-species SAMap pipeline (Phase 0.1)
atarashansky
committed
refactor: switch SAMap imports from samalg to vendored samap.sam
atarashansky
committed
feat(backend): add Backend dispatch class for numpy/scipy <-> cupy/cupyx (Phase 2.1)
atarashansky
committed
refactor(core): split mapping.py into focused modules (Phase 1)
atarashansky
committed
feat(expand): add BFS-based _smart_expand with numba kernel (Phase 3.3)
atarashansky
committed
feat(backend): wire Backend into SAMAP + replace non-hot-path LIL (Phase 2.2/2.3)
atarashansky
committed
feat(coarsening): streaming per-pair mutual-NN construction (Phase 3.2)
atarashansky
committed
feat(projection): precomposed feature translation — eliminate N×G Xtr intermediate (Phase 3.1)
atarashansky
committed
feat(orchestration): wire projection precompute into _Samap_Iter loop (Phase 3 finalization)
atarashansky
committed
feat(correlation): batched on-demand Xavg + dict-free kernel (Phase 3.4)
atarashansky
committed
feat(pca): add randomized SVD with implicit centering (Phase 4.4)
atarashansky
committed
feat(knn): GPU/CPU kNN dispatch via approximate_knn (Phase 4.2)
atarashansky
committed
feat(correlation): GPU-ready _replace via vectorised dispatch (Phase 4.3)
atarashansky
committed
feat(bench): phase-level legacy-vs-optimized benchmark suite (Phase 5.3)
atarashansky
committed
release: flip optimization defaults to ON, regen golden, v3.0.0 (Phase 5 finalization)
atarashansky
committed
feat(correlation): auto-select batched vs materialised by estimated memory
atarashansky
committed
lint
atarashansky
committed