3 packages tagged with “dimensionality-reduction”
Pure, unoptimized UMAP implementation for Flowthru - direct port from Python reference implementation by Leland McInnes
Production-ready UMAP library with revolutionary HNSW optimization (50-2000x faster), spectral initialization (default), and InitializationMethod enum for explicit control. Features: 1M+ dataset support with 80-85% memory reduction, 5-level outlier detection for AI/ML validation, arbitrary dimensions (1D-50D), multiple distance metrics (Euclidean, Cosine, Manhattan, Correlation, Hamming), enhanced progress reporting, complete model persistence, and comprehensive hyperparameter control (localConnectivity, bandwidth). Perfect for production AI pipelines requiring high-quality embeddings at scale.
Production-ready PACMAP library with 66% smaller persistence files, 3.1-12.5x performance optimization, HNSW acceleration, and cross-platform 64-bit binaries. Features: PACMAP (Pairwise Controlled Manifold Approximation and Projection) algorithm with OpenMP 8-thread parallelization, AVX2/AVX512 SIMD optimization, HNSW (Hierarchical Navigable Small World) acceleration, optimized model persistence (v2.8.32), advanced quantization (60% storage savings), comprehensive progress reporting, model persistence, and cross-platform deployment. Perfect for production AI pipelines requiring high-performance dimensionality reduction with enterprise-grade stability and efficient storage.