NEWS

---

bigKNN 0.3.0 (2026-04-01)

• Added execution plans with memory-budget-based block sizing and thread hints.
• Added serializable prepared references with knn_load_prepared() and knn_validate_prepared().
• Added exact radius-graph helpers, streamed directed kNN graph jobs, checkpointed radius streaming, and resumable checkpoints.
• Added result coercion helpers plus exact recall summaries and candidate reranking utilities.

bigKNN 0.2.0

• Added prepared exact references with knn_prepare_bigmatrix(), knn_search_prepared(), and knn_search_stream_prepared() for repeated search workloads.
• Added exact radius-search APIs with in-memory and streamed outputs, including count_within_radius_bigmatrix(), radius_bigmatrix(), and radius_stream_bigmatrix().
• Expanded metric support from Euclidean distance to "sqeuclidean" and "cosine".
• Added exact graph builders for directed kNN, mutual kNN, and shared-nearest-neighbour graphs.

bigKNN 0.1.0

• Initial release with exact Euclidean k-nearest-neighbour search on bigmemory::big.matrix references.
• Added self-search and external-query search against big.matrix and dense matrix inputs.
• Added both in-memory and streamed neighbour outputs through knn_bigmatrix() and knn_stream_bigmatrix().
• Established package documentation, compiled backend registration, and the first regression tests.