spectral-library 0.3.0¶

Released 2026-04-04.

Highlights¶

Rust-accelerated batch mapping via a PyO3/Rayon native extension for spectrum reconstruction, target-sensor projection, segment merging, and neighbor refinement
LinearSpectralMapper for fixed input @ weights + bias high-throughput mapping without per-sample KNN retrieval
Dense batch vectorization across the full segment retrieval, reconstruction, and projection pipeline
Cross-platform compiled wheels (Linux x86_64/aarch64, macOS x86_64/arm64, Windows AMD64) built via cibuildwheel

What Changed¶

New Public API¶

Symbol	Purpose
`LinearSpectralMapper`	Compiled linear mapper for million-pixel throughput
`BatchMappingArrayResult`	Dense result type for array-oriented batch mapping
`SpectralMapper.compile_linear_mapper()`	Compile a prepared runtime into a reusable linear mapper
`SpectralMapper.map_reflectance_debug()`	Single-sample mapping with full diagnostic payload
`SpectralMapper.map_reflectance_batch_debug()`	Batch mapping returning rich `MappingResult` objects
`SpectralMapper.map_reflectance_batch_arrays()`	Explicit dense batch path returning `BatchMappingArrayResult`
`SpectralMapper.map_reflectance_batch_ndarray()`	Direct ndarray input batch mapping
`SpectralMapper.map_reflectance_batch_to_zarr()`	Stream large batch results to Zarr stores

Slim vs Debug Paths¶

map_reflectance() and map_reflectance_batch() now default to slim result paths that skip neighbor ids, per-segment diagnostics, and review payloads. Use the _debug() variants when you need the full payload.

CLI Additions¶

--output-format zarr and --output-chunk-size flags for streaming batch output

Build and Distribution¶

Wheels now include a compiled Rust extension for all major platforms
cibuildwheel builds in the release workflow replace pure-Python wheel builds
Package checks CI installs the Rust toolchain for editable-install tests

Full Changelog¶

See CHANGELOG.md for the complete list of additions, changes, and fixes.