Skip to content

Infer folder hierarchies, distribute CUDA fits, and expand tutorial gallery - #186

Merged
jeremymanning merged 1 commit into
mainfrom
codex/hierarchy-gpu-tutorials
Oct 29, 2026
Merged

jeremymanning merged 1 commit into
mainfrom
codex/hierarchy-gpu-tutorials

Conversation

@jeremymanning

@jeremymanning jeremymanning commented Oct 29, 2026 •

Copy link
Copy Markdown
Member

Nested NIfTI/image folders infer hierarchical templates using folder names, and CUDA subject fits run concurrently across selected GPUs. The quick-start guide and all 13 tutorials explain and illustrate the API; each tutorial has a gallery card linking directly to its notebook. The 2D image example reaches finer detail with up to 320 factors, and the PyPI badge appears only in README.

PR #185 is merged before this branch begins, so this diff contains only the follow-up work.

Fresh independent reviewers identify and resolve ambiguous folder ancestors, empty group-tree outputs, excluded-GPU initialization, complex-start validation, optional notebook imports, clipped markers, and stale remote animation previews. Animation tutorials now embed the files they generate. A validation runner check requires visual output from every tutorial.

Download the rendered documentation preview.

Validation:

  • Installed wheel: 1,119 passed, 19 skipped, 28 deselected.
  • All 13 notebooks execute with visual output; 69 figures and representative animation frames reviewed. Changed outputs are refreshed after executable-code checks.
  • All 341 baseline/candidate fingerprint arrays are bit-identical.
  • Network checks: 28 passed, 2 skipped across installed-package checks and source-based installer checks. An initial copied-test harness omitted package sources needed by installer subprocesses; the corrected installer run passes all 10 tests.
  • Source-distribution smoke/statistics tests: 15 passed.
  • Ruff, formatting, mypy, Twine, explicit PyPI README rendering, and strict Sphinx build pass; all nine gallery examples execute.
  • Gallery thumbnails and titles link directly to all 13 tutorials.

Actual MPS tests pass. CUDA hardware is unavailable locally: real multi-device and excluded-device tests are included but skip here; queue routing is exercised with real NumPy fits. Fitting supports unequal-depth group paths and repeated leaf names, while permutation inference explicitly requires equal-depth paths with distinct subgroup labels under different parents.

Final commit checks: tests, installed-wheel/tutorial matrix, BrainIAK, and documentation. All final commit checks pass: all 16 source-test jobs, all 16 installed-wheel/tutorial jobs, BrainIAK, and documentation. The GitHub package release remains a draft; no PyPI/TestPyPI upload or Zenodo action is authorized. See notes/RELEASE_REVIEW.md for evidence and final-review gates.

@jeremymanning
jeremymanning marked this pull request as ready for review October 29, 2026 17:13
@jeremymanning
jeremymanning merged commit 22eff8d into main Oct 29, 2026
34 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant