fix: optimal threshold computation

[michele-milesi]

Description

Provide a summary of the modification as well as the issue that has been resolved.

• Compute Optimal Threshold with single target with combined method: fixed epsilon and torch.nextafter

[Copilot]

Pull request overview

This pull request fixes the optimal threshold computation in the OptimalF1 metric for the edge case when only a single target class is present. The fix uses a combined approach of torch.nextafter and a fixed epsilon value to ensure numerical correctness across different floating-point precisions.

Changes:

• Enhanced threshold computation for single-target scenarios using torch.nextafter combined with fixed epsilon
• Added validation to detect and raise an error for infinite threshold values
• Updated version to 0.7.0.dev151 and added changelog entry

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 5 comments.

File	Description
src/anomalib/utils/metrics/optimal_f1.py	Improved threshold computation with torch.nextafter and epsilon combination for single-target case, with infinity check
pyproject.toml	Version bump to 0.7.0.dev151
CHANGELOG.md	Added changelog entry for the fix

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

The optimal_f1_score tensor is created without specifying dtype or device, which may lead to inconsistency with other tensors in the computation. Consider creating it with the same dtype and device as current_preds for consistency: torch.tensor(1.0, dtype=current_preds.dtype, device=current_preds.device).

Suggested change

	optimal_f1_score = torch.tensor(1.0)
	optimal_f1_score = torch.tensor(1.0, dtype=current_preds.dtype, device=current_preds.device)