pyGAPIT 2.0.0

Added

• Finalize the 2.0 storage, streaming-ingestion, HoloViews visualization, and
  scientific-parity features described in the two release candidates below.

Changed

• Promote 2.0.0rc2 to the stable 2.0 release without further runtime changes
  after the complete CI, package, supported-platform, optional-feature, and
  GAPIT 3.5 parity gates passed.
• Require Python 3.12 or newer and support Windows x86-64, Linux x86-64 and
  ARM64, and macOS on Apple Silicon.

---

Full Changelog: v2.0.0rc2...v2.0.0

pyGAPIT 2.0.0rc2

Changed

• Support macOS on Apple Silicon only; remove the Intel macOS platform because
  the required Numba/llvmlite stack is unavailable for the supported Python
  versions.
• Run supported-platform smoke tests through the locked Pixi environments and
  keep R-free installation checks isolated from the GAPIT parity environment.

Security

• Pin GitHub Actions to immutable commit SHAs, let Dependabot maintain those
  pins, minimize publication permissions, and gate the CI matrix behind a
  single required status check.

Maintenance

• Add structured issue forms, a pull-request template, and a security policy
  for external reports and contributions.

---

Full Changelog: v2.0.0rc1...v2.0.0rc2

pyGAPIT 2.0.0rc1

Added

• Typed Visualization objects with independently assignable notebook
  backends, direct native render(), and format-aware saving. Static formats
  select Matplotlib automatically; 3D PCA excludes Bokeh.
• Backend-neutral HoloViews objects for Manhattan, QQ, kinship, PCA, genomic
  selection, and phenotype plots, with rendering and saving delegated to
  HoloViews' Matplotlib, Bokeh, and Plotly backends, built on shared,
  backend-independent Manhattan data and genomic-axis preparation.
• A reproducible benchmark for genomic-axis and complete Manhattan data
  preparation, HoloViews object construction, and opt-in Matplotlib, Bokeh, or
  Plotly rendering at configurable marker counts.
• An optional backend using the Zarr Python 3.x API and Zarr format 2 storage,
  with the same labeled, chunk-readable contract as NumPy and HDF5.
• Chunked conversion between labeled NumPy, HDF5, and Zarr genotype stores
  without materializing the complete genotype matrix.
• Direct bounded sample-row import from numeric GD/GM files into any
  genotype-store backend without rescanning the source for every marker block.
• Single-pass marker-block import from HapMap files into any genotype-store
  backend.
• A reproducible large-marker benchmark over bounded numeric import, VanRaden
  kinship, PCA, a GLM marker scan, Manhattan preparation, and rendering at
  100k, 1M, and 10M markers with selectable NumPy, HDF5, and Zarr store
  backends, reporting wall time and process peak RSS.
• A staged numeric-import profiler that separates source and metadata setup,
  TSV parsing plus imputation, and NumPy persistence.

Changed

• Document the Python 3.12 and HoloViews visualization migration from 1.2.4.
• Reuse precomputed genomic ordering when building GWAS tables and bypass
  DataFrame sorting and merging when multi-model results already share marker
  order.
• Store Manhattan hover fields as HoloViews value dimensions instead of
  allocating backend-specific formatted labels.
• Replace backend-specific plotting arguments and native figure return values
  with a typed HoloViews visualization view; construction selects notebook
  display preference while render() and save_plot() handle output.
• Use a linear, low-allocation genomic-axis path when chromosome blocks are
  contiguous while preserving arbitrary-order input compatibility.
• Reuse zero-copy chromosome slices while constructing exact Manhattan points
  and combine alternating chromosome colors in one categorical Datashader
  raster.
• Require Python 3.12 or newer and test the complete optional-storage stack on
  Python 3.12 and 3.14.
• Extend disk-backed top-level workflows through SUPER selection and sBLUP
  prediction while materializing and reusing only the bounded pseudo-QTN
  candidate pool.
• Support bounded disk-backed marker scans in direct and top-level CMLM.
• Extend direct and top-level MLMM to disk-backed marker batches while reading
  only its selected cofactors into each fixed-effect design.
• Support disk-backed FarmCPU iterations through the shared bounded GLM scan,
  reward substitution, and pseudo-QTN kinship paths.
• Add disk-backed BLINK with bounded GLM and LD-pruning reads, plus reuse of its
  truncated BIC candidate pool.
• Support disk-backed cBLUP prediction by streaming VanRaden marker blocks;
  compression still operates on the complete in-memory kinship matrix.

Removed

• Backend-native plotting functions are removed rather than deprecated, with
  no compatibility wrappers:

  • manhattan_plot() → manhattan() plus save_plot()
  • manhattan_interactive() → manhattan(..., backend="bokeh" | "plotly")
  • pca_plot_3d_interactive() → pca_plot_3d()

• The plotting save_path= parameters, such as qq_plot(..., save_path=) and
  kinship_heatmap(..., save_path=). Construct a Visualization and call
  save_plot() instead. MIGRATING.md records the full mapping.

Fixed

• Failed NumPy, HDF5, and Zarr writes remove the incomplete genotype store so
  the same destination can be retried safely.
• Failed plot generation no longer reports stale files from an earlier run as
  outputs produced by the current analysis.
• Static-bin FarmCPU no longer derives a pseudo-kinship REML variance fit.
  GAPIT's static-bin path returns no variance components, so the result fields
  vg, ve, and h2 are now fixed at 0.0 instead of an unsupported
  estimate.
• sBLUP PEV for intentionally low-rank pseudo-kinship now inverts the kinship
  at GAPIT's MASS::ginv tolerance instead of NumPy's smaller default cutoff,
  so PEV and prediction-error values match the R reference.
• MLMM reproduces GAPIT's complete forward and backward cofactor paths, so the
  retained model and its marker statistics match the R reference at every
  iteration.

---

Full Changelog: v1.2.4...v2.0.0rc1

pyGAPIT 1.2.4

Full Changelog: v1.2.3...v1.2.4

pyGAPIT 1.2.3

Full Changelog: v1.2.2...v1.2.3

pyGAPIT 1.2.2

Full Changelog: v1.2.1...v1.2.2

pyGAPIT 1.2.1

• Publish the distribution as pygapit-ng while retaining pygapit as the
  import package and command-line entry point.