Fix unused imports, min_redundant_inds remainder bug, and ClassBalancedSampler index mapping

[Copilot]

Addresses several correctness and lint issues flagged in code review on the rewrite PR.

Unused imports

Removed F401 violations across test files:

• import json from test_dataset.py
• import pytest from test_geometry.py, test_sampler.py
• import tempfile, pathlib, import torch from test_multidataset.py
• import zarr as z from test_image.py

min_redundant_inds remainder bug

When replacement=False and k > n, the previous implementation silently dropped indices when k % n != 0:

# Before: k=7, n=3 → returned 6 indices, not 7
return torch.cat([torch.randperm(n, generator=rng) for _ in range(k // n)])

# After: always returns exactly k indices
full_perms = k // n
remainder = k % n
parts = [torch.randperm(n, generator=rng) for _ in range(full_perms)]
if remainder > 0:
    parts.append(torch.randperm(n, generator=rng)[:remainder])
return torch.cat(parts)

ClassBalancedSampler fixes

Empty active classes guard — __init__ now raises ValueError immediately if the crop-class matrix has no annotated classes, rather than crashing with a cryptic error on active_counts.min() at iteration time.

Correct sample index mapping — __iter__ previously yielded crop_idx directly, which is a row in the crop-class matrix (one row per dataset/sub-dataset), not a sample index. This meant only indices [0, n_crops) were ever yielded, ignoring the vast majority of samples. Now maps crop rows to actual sample indices:

• n_crops == len(dataset): direct mapping (no-op, preserves existing behavior)
• Dataset has datasets + cumulative_sizes (CellMapMultiDataset / ConcatDataset): samples randomly within the matched sub-dataset's global index range
• Generic fallback: partitions [0, len(dataset)) into n_crops contiguous segments and samples within the crop's segment

---

✨ Let Copilot coding agent set things up for you — coding agent works faster and does higher quality work when set up for your repo.

[codecov]

Codecov Report

❌ Patch coverage is 57.14286% with 15 lines in your changes missing coverage. Please review.
✅ Project coverage is 80.97%. Comparing base (c7141ce) to head (4448a56).
⚠️ Report is 9 commits behind head on rewrite.

Files with missing lines	Patch %	Lines
src/cellmap_data/sampler.py	48.27%	15 Missing ⚠️

Additional details and impacted files

@@             Coverage Diff             @@
##           rewrite      #70      +/-   ##
===========================================
- Coverage    81.36%   80.97%   -0.40%     
===========================================
  Files           23       23              
  Lines         1567     1603      +36     
===========================================
+ Hits          1275     1298      +23     
- Misses         292      305      +13     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[Copilot]

Pull request overview

This PR addresses lint/correctness issues in CellMap-Data’s sampling utilities and cleans up unused test imports, improving both reliability and code hygiene.

Changes:

• Remove unused imports in multiple test modules to resolve F401 lint violations.
• Fix min_redundant_inds to always return exactly k indices when replacement=False and k > n (including remainder handling).
• Fix ClassBalancedSampler by adding an early “no active classes” guard and mapping crop-matrix rows to actual dataset sample indices during iteration.

Reviewed changes

Copilot reviewed 7 out of 7 changed files in this pull request and generated 5 comments.

Show a summary per file

File	Description
tests/test_sampler.py	Removes unused pytest import.
tests/test_multidataset.py	Removes unused imports (torch, tempfile, pathlib).
tests/test_image.py	Removes unused local zarr import.
tests/test_geometry.py	Removes unused pytest import.
tests/test_dataset.py	Removes unused local json import.
src/cellmap_data/utils/misc.py	Fixes remainder handling in min_redundant_inds for k > n without replacement.
src/cellmap_data/sampler.py	Adds empty-active-class guard and corrects sampler output to yield dataset sample indices instead of crop-matrix row indices.

[Copilot]

The sampler now yields a mapped sample_idx, but the surrounding terminology still calls this a “crop index” (and the class docstring describes yielding “that crop index”). This makes the API/behavior confusing for callers and future maintainers. Consider renaming crop_idx to something like row_idx (matrix row) and updating the docstring/comments to reflect that __iter__ yields dataset sample indices (global indices for Concat/MultiDataset).

[Copilot]

Pull request overview

Copilot reviewed 7 out of 7 changed files in this pull request and generated 2 comments.

[Copilot]

In the generic fallback mapping, when len(self.dataset) == 0 this branch sets end = max(total, 1) and can yield sample_idx = 0, which is an invalid index for an empty dataset (and will likely crash downstream when DataLoader tries to fetch it). Consider explicitly erroring out when total == 0 (or forcing samples_per_epoch to 0) instead of manufacturing a [0, 1) range.

Suggested change

	if self.n_crops <= 1 or total <= 0:
	start, end = 0, max(total, 1)
	if total <= 0:
	raise ValueError(
	"ClassBalancedSampler: cannot sample from an empty dataset "
	"in generic fallback mapping (len(dataset) == 0). "
	"Consider setting samples_per_epoch=0 or removing the "
	"sampler for this split."
	)
	if self.n_crops <= 1:
	start, end = 0, total

[Copilot]

PR description says import pytest was removed from tests/test_sampler.py, but this file still imports (and now uses) pytest in the newly added test_raises_when_no_active_classes. Consider updating the PR description to avoid confusion when reviewing the unused-import cleanup.