Fix RandGridPatch whole-dimension patch sizes

[Rajioba1]

Fixes#9046.

Description

RandGridPatch documents that patch_size entries of 0 or None select the whole dimension. When max_offset is omitted, the transform currently computes the default offset range with s % p, which raises for 0 and None patch dimensions.

This PR treats whole-dimension patch entries as having zero random offset for that dimension, matching the documented behavior and GridPatch behavior. It adds regression coverage for both RandGridPatch and RandGridPatchd with patch_size=(0, 2) and patch_size=(None, 2).

Types of changes

• Non-breaking change (fix or new feature that would not break existing functionality).
• Breaking change (fix or new feature that would cause existing functionality to change).
• New tests added to cover the changes.
• Integration tests passed locally by running ./runtests.sh -f -u --net --coverage.
• Quick tests passed locally by running ./runtests.sh --quick --unittests --disttests.
• In-line docstrings updated.
• Documentation updated, tested make html command in the docs/ folder.

Local validation:

• python -m tests.transforms.spatial.test_rand_grid_patch
• python -m tests.transforms.spatial.test_rand_grid_patchd
• git diff --check

Notes:

• black and isort were not installed in my local Python environment, so I could not run those checks directly.
• python -m ruff check monai/transforms/spatial/array.py tests/transforms/spatial/test_rand_grid_patch.py tests/transforms/spatial/test_rand_grid_patchd.py reports an unrelated pre-existing UP038 warning in array.py outside this PR's changed lines.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: 8181fe72-9912-472c-9a68-21d23e4b9c12

📥 Commits

Reviewing files that changed from the base of the PR and between 7e9303e and 2ce2116.

📒 Files selected for processing (2)

• tests/transforms/spatial/test_rand_grid_patch.py
• tests/transforms/spatial/test_rand_grid_patchd.py

💤 Files with no reviewable changes (2)

• tests/transforms/spatial/test_rand_grid_patchd.py
• tests/transforms/spatial/test_rand_grid_patch.py

---

📝 Walkthrough

Walkthrough

RandGridPatch.randomize now handles zero patch dimensions without modulo-by-zero errors. New parameterized tests verify zero-dimension patches for RandGridPatch and RandGridPatchd across supported input types.

Estimated code review effort: 2 (Simple) | ~10 minutes

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Linked Issues check	⚠️ Warning	The changes address zero-sized dimensions, but the summary does not confirm support or regression coverage for the required None dimensions in issue #9046.	Confirm that default max_offset handles patch_size=None and add or verify regression tests for both RandGridPatch and RandGridPatchd.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly describes the main change: fixing whole-dimension patch sizes in RandGridPatch.
Description check	✅ Passed	The description explains the bug, solution, scope, tests, and local validation, with only optional checks left unreported.
Out of Scope Changes check	✅ Passed	All reported code and test changes relate directly to the RandGridPatch whole-dimension behavior described in issue #9046.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[coderabbitai]

🧹 Nitpick comments (1)

monai/transforms/spatial/array.py (1)

3570-3570: 🎯 Functional Correctness | 🔵 Trivial | ⚡ Quick win

Make the dimension contract explicit.

zip silently truncates when array.shape[1:] and self.patch_size have different lengths. This can create an incomplete max_offset and an invalid self.offset. Pass strict=True.

Proposed fix

- max_offset = tuple(s % p if p else 0 for s, p in zip(array.shape[1:], self.patch_size))+ max_offset = tuple(+ s % p if p else 0+ for s, p in zip(array.shape[1:], self.patch_size, strict=True)+ )

Ruff B905 identifies this call because it omits strict=.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@monai/transforms/spatial/array.py` at line 3570, Update the max_offset
computation in the relevant transform method to call zip with strict=True,
ensuring array.shape[1:] and self.patch_size must have matching dimensions
instead of being silently truncated.

Source: Linters/SAST tools

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Nitpick comments:
In `@monai/transforms/spatial/array.py`:
- Line 3570: Update the max_offset computation in the relevant transform method
to call zip with strict=True, ensuring array.shape[1:] and self.patch_size must
have matching dimensions instead of being silently truncated.

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: 1087d4fe-91b6-4634-bbc4-e4bffa33fc5a

📥 Commits

Reviewing files that changed from the base of the PR and between 8690ae7 and 7e9303e.

📒 Files selected for processing (3)

• monai/transforms/spatial/array.py
• tests/transforms/spatial/test_rand_grid_patch.py
• tests/transforms/spatial/test_rand_grid_patchd.py

[vikashg]

Review

Effective diff against current dev: 1 source line + 2 test lines (the MetaTensor.astype and write_metrics_reports changes were already merged via #8911 and #8902).

The RandGridPatch fix

The one-line change is correct and minimal:

# Before (crashes for p=0 or p=None):max_offset=tuple(s%pfors, pinzip(array.shape[1:], self.patch_size))
# After:max_offset=tuple(s%pifpelse0fors, pinzip(array.shape[1:], self.patch_size))

I confirmed:

• ZeroDivisionError on current dev with patch_size=(0, 2) ✅
• Fix resolves it for both 0 and None (whole-dimension) entries ✅
• Normal patch_size=(2, 2) still works (regression pass) ✅
• Edge case patch_size=(0, 0) returns full image as single patch ✅
• All 62 tests in test_rand_grid_patch.py pass ✅

Suggestions

1. Add a patch_size=(None, 2) test case alongside the (0, 2) case. The docs say both 0 and None select the whole dimension — covering both in tests makes the contract explicit.

2. (Nit) The PR description mentions this fixes three things, but fix(MetaTensor): astype with torch dtype now returns MetaTensor preserving metadata #8911 and Fix class_labels mutation across multi-metric write_metrics_reports #8902 are already in dev. Consider rebasing so the PR history only shows the RandGridPatch commits.

Verdict

Approve — clean, minimal, well-tested fix for a documented bug. The nit about adding a None test case is optional.

[Rajioba1]

Thanks for the thorough review, @vikashg — really appreciate it.

Good call on covering None explicitly. I initially had a patch_size=(None, 2) case, but removed it because it exposed a separate metadata issue rather than just exercising this fix.

Unlike 0, None makes np.array(self.patch_size) object dtype:

np.array((None, 2)).dtype# dtype("O")

That object-dtype spatial_shape flows into the patch metadata, and through the MetaTensor test path it can produce object-dtype metadata that breaks downstream collation in the shared min-dep test run.

I see two clean options:

1. Add a patch_size=(None, 2) test scoped to non-MetaTensor inputs in this PR, so the documented behavior is covered without expanding scope.
2. Normalize None/0 patch-size entries to the actual dimension size before writing spatial_shape, so the MetaTensor case is also safe. That's stronger, but slightly broader and may be better as a follow-up.

Happy to do either — which would you prefer?