Changelog

Changelog#

All notable changes to this project are documented here. Format follows Keep a Changelog.

0.2.2 — 2026-05-06#

Added#

  • instrument_primer.ipynb — standalone notebook explaining NIRISS GR150R/F150W grism geometry: sensitivity curves, trace geometry and footprint table, dispersion solution, single-source and multi-source (10-source aggregate) per-order images, and a full-stamp source coverage map. Integrated into the Sphinx documentation under a new Background section.

  • Sphinx API documentation — full autodoc coverage across five RST pages (modules, instrument_basis, operators, solvers, overview); four Graphviz architecture diagrams; all cross-references resolve; zero new -W warnings.

  • FISTA adaptive restart — JAXProximalSolver gains tol (convergence tolerance), restart (gradient-condition adaptive momentum reset), and callback parameters.

  • Comparison notebooks — pre-executed with outputs committed:

    • comparison_solver_accuracy.ipynb: LSQR vs FISTA on a crowded scene + RMSE sweep over source density.

    • comparison_computational.ipynb: memory, runtime, and convergence benchmarks.

0.2.0 — 2026-05-05#

Added#

  • JAXOperator — JAX-based forward operator with compact trace representation (trace_indices[K,O,L] int32 + weights[O,L,M] float32); out-of-bounds wavelengths absorbed by a ghost pixel at n_pix rather than masked at call time.

  • JAXProximalSolver — FISTA with group-L1 proximal operator (group_soft_threshold) and power_iteration for Lipschitz constant estimation.

  • Both classes exposed in the public API (spectrex.JAXOperator, spectrex.JAXProximalSolver).

  • NoiseModel.sample() for mock Poisson + read-noise generation.

  • analysis_rmse_vs_density.ipynb — RMSE vs source density sweep, pre-executed.

  • mock_example.ipynb — end-to-end Phase 1 worked example with noiseless and noisy recovery.

  • Package-level logger with NullHandler (no output unless the caller configures logging).

  • Slow integration test (@pytest.mark.slow) for the full JAX pipeline with real testdata.

0.1.0 — 2026-05-05#

First installable release. The original codebase was a collection of standalone scripts for NIRISS WFSS spectral extraction — no package structure, no tests, no documentation.

Added#

  • src/spectrex/ package layout with pyproject.toml / uv toolchain.

  • EigenspectraBasis — frozen dataclass; loads PCA components from CSV; read-only array views; reconstruct() and broadband_image().

  • InstrumentConfig — loads GR150R calibration files; get_trace() returns wavelength-resolved pixel positions; per-order sensitivity arrays.

  • ForwardOperatorProtocol — @runtime_checkable typing Protocol defining the apply() / apply_transpose() interface.

  • SciPySparseOperator — CSR-backed implementation of the protocol; explicit sparse matrix construction from trace geometry.

  • SpectralSolver — LSQR / LSMR with optional Tikhonov regularisation.

  • NoiseModel — Poisson + Gaussian read noise model.

  • pytest suite: fast unit tests + slow integration tests against real testdata/; slow tests skipped in default CI via @pytest.mark.slow.

  • Sphinx documentation scaffold.