v0.1.2

[0.1.2] — 2026-04-29

Changed

• Generalized trait associations — trait_columns is now fully data-driven. Any
  column in the sample table can be listed as a trait: numeric columns use Pearson
  correlation and binary categorical columns use a Welch t-test. Previously, Dx and
  Age were hardcoded and required to be present. #2
• Shared compute_trait_associations utility — duplicated _trait_associations
  logic has been consolidated into a single function in models/base.py. All backends
  (baseline, latent, graph, VAE, GPU-latent, WGCNA) now delegate to this shared
  implementation. #2
• Default trait_columns updated — the built-in default changed from
  ["Dx", "Age"] to ["Age"] to avoid assuming a diagnosis column is present. #2

CI

• Added Python 3.14 to the test matrix.
• Removed the separate test-torch CI job; PyTorch-gated tests are now handled by
  fixture skipping in the main matrix.

---

[0.1.1] — 2026-04-25

Added

• eigengene_table in FitArtifacts — all backends now compute and return a
  per-module eigengene matrix (modules × samples) alongside the trait association table. #1
• Snapshot export — isograph freeze-real / isograph benchmark now writes
  eigengene_table.parquet to the output directory when at least one module is detected. #1
• CHANGELOG.md and PyPI installation documentation.

Fixed

• Corrected documentation for the WGCNA backend: rpy2 is not required; the
  backend calls Rscript directly and expects R and the WGCNA package in PATH.

Full Changelog: v0.1.1...v0.1.2