release: prepare statgpu 0.2.4

[TheHiddenObserver]

Summary

Prepare the next PyPI and GitHub release as statgpu 0.2.4 after merging PR #87.

This is a focused release-only pull request. It does not change estimator, solver, backend, inference, or numerical implementation code.

Current status: RELEASE_READY_ALL_GATES_PASS.

• current head: 31a00cb00d1561f279e78764aa6da307cc9cbc66;
• diff is limited to six release-facing files;
• all exact-head hosted gates pass;
• exact-head physical NVIDIA acceptance passes on both Torch and CuPy;
• merge by squash only, then tag the release-PR merge commit as v0.2.4.

Changes

• bump project.version and statgpu.__version__ from 0.2.3 to 0.2.4;
• add the authoritative user-facing GitHub Release document at .github/releases/v0.2.4.md;
• convert the root, English, and Chinese Unreleased material into synchronized 0.2.4 — 2026-08-06 release sections;
• summarize the user-visible LogisticRegression, GLM, CV, inference, solver, backend, sklearn-compatibility, import-order, and Torch compile changes delivered by PR fix: harden runtime and estimator contracts #87;
• document behavioral changes, installation commands, platform support, upgrade implications, known limits, validation evidence, and release-history links.

Release positioning

Version 0.2.4 is a patch release. PR #87 materially corrects existing public estimator and runtime contracts but does not introduce a new model family requiring a minor-version increment.

Exact-head hosted validation

Validated on 31a00cb00d1561f279e78764aa6da307cc9cbc66:

• Tests #1402: success
  • complete CPU suite: 2239 passed, 719 skipped, 20 warnings in 41.35s;
  • static contracts: success;
  • documentation contracts: success;
  • Python 3.9, 3.10, 3.11, and 3.12 regression jobs: success;
• Maintenance compatibility #415: success under scikit-learn 1.2.2, 1.3.2, and latest;
• Release notes validation #428: success;
• Release package validation #434: success
  • version declarations, wheel/sdist build, twine check, artifact-content checks, and clean sdist installation: success;
  • Ubuntu, Windows, and macOS installation of the same universal wheel artifact: success.

Exact-head physical GPU acceptance

Validated on release head 31a00cb00d1561f279e78764aa6da307cc9cbc66.

Torch / RTX 4090

Environment: px2-jcy, RTX 4090 (compute capability 8.9), PyTorch 2.8.0+cu128.

• Torch compile + CUDA Graph pytest matrix: 9 passed in 8.97s;
• IRLS + LogisticRegression runtime contracts: all assertions passed;
• IRLS parameters: [0.0, 1.0];
• prediction dtype: torch.int64;
• weighted logistic log-likelihood: -0.557463468137747.

CuPy / Tesla P100

Environment: hz-4, Tesla P100-SXM2-16GB, CuPy 13.6.0.

• IRLS + LogisticRegression runtime contracts: all assertions passed;
• IRLS parameters: [0.0, 1.0];
• prediction dtype: int64;
• weighted logistic log-likelihood: -0.5574634681377472.

The two backends agree within floating-point precision, converge to the same IRLS parameters, and satisfy the required integer prediction-dtype contract.

Merge and publication

1. Merge by squash only.
2. Confirm the repository PYPI_TOKEN remains valid and project-scoped.
3. Tag the release-PR merge commit as v0.2.4.
4. Push the tag to trigger the existing PyPI and GitHub Release publication workflow.
5. Verify pip install --no-cache-dir statgpu==0.2.4, the published wheel/sdist, and the GitHub Release body.

[cursor]

Bugbot is not enabled for your account, so this pull request was not reviewed.

Enable Bugbot in the Cursor dashboard to get automatic reviews on future PRs.

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.
To continue using code reviews, you can upgrade your account or add credits to your account and enable them for code reviews in your settings.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[TheHiddenObserver]

Post-merge TestPyPI acceptance

The v0.2.4 distributions were built and uploaded to TestPyPI from the validated release source, and the installed TestPyPI wheel completed the release smoke matrix successfully.

• wheel: statgpu-0.2.4-py3-none-any.whl
• contents: 247 files, approximately 1,651 KB
• functional smoke result: 10/10 passed

Validated public model paths:

• LinearRegression: fit + predict
• Lasso: fit
• Ridge: fit
• ElasticNet: fit
• PoissonRegression: fit + predict
• PenalizedLinearRegression with SCAD: fit + predict
• PenalizedLinearRegression with MCP: fit + predict
• LogisticRegression: binary fit + predict
• CoxPH: survival fit
• GroupLasso: group-penalty fit

Together with the exact-head hosted and physical-GPU acceptance already recorded in this PR, this closes the pre-publication package rehearsal for 0.2.4.

Release PR merge commit to tag: 0aeeb95b60e3e274053b8f1b6427ae50c8eec015.

[TheHiddenObserver]

v0.2.4 publication complete

The tagged release workflow completed successfully for v0.2.4 on merge commit 0aeeb95b60e3e274053b8f1b6427ae50c8eec015.

• workflow run: 31083940686 — success;
• tag/package/release-note verification: passed;
• pure-Python wheel and sdist build: passed;
• twine check: passed;
• clean release-wheel smoke installation: passed;
• PyPI upload job: passed;
• GitHub Release publication job: passed;
• retained workflow artifact: statgpu-tagged-release-distributions;
• GitHub Release title: statgpu 0.2.4.

Published GitHub Release assets:

• statgpu-0.2.4-py3-none-any.whl
  • SHA-256: 297c1b296a89c37baf180281727db97a6c56b7b76ae4bbfcf26e63e39b669e3c
• statgpu-0.2.4.tar.gz
  • SHA-256: c43c8f63f9198c208881383e1e2023a012b693def171ec3fb2d791a61e3dc7ac

Issue bookkeeping is also complete: #45, #81, and #82 were closed as completed; #83 was closed as not planned after recording the disposition of the unavailable historical diagnostic reproducer. The repository currently has no open issues.