Fix size attribute error for precision/recall/f1

[Maxwell-Jia]

What does this PR do?

Fix #655 .
Fix AttributeError in precision/recall/f1 metrics when handling scalar outputs from scikit-learn 1.6.0.

Description

With the release of scikit-learn 1.6.0, some metric functions (e.g., precision_score, recall_score, f1_score) may return float values instead of numpy arrays for single-value results. The current implementation in evaluate assumes the presence of a size attribute for all outputs, which causes an AttributeError when handling scalar outputs.

This PR modifies the return statement to safely handle both numpy arrays and scalar outputs using getattr(score, 'size', 1), making the metrics compatible with both scikit-learn 1.6.0 and earlier versions.

Changes

Modified return statements in three metrics:

• metrics/precision/precision.py
• metrics/recall/recall.py
• metrics/f1/f1.py

Changed from:

return {"metric_name": float(score) if score.size == 1 else score}

to:

return {"metric_name": score if getattr(score, 'size', 1) > 1 else float(score)}

[Maxwell-Jia]

Hi @albertvillanova! I noticed you've recently reviewed some PRs in this area. I hope it's okay to bring this to your attention - this PR addresses the scikit-learn 1.6.0 compatibility issue (from #655) that's affecting several users. Would really appreciate your insights when you have a moment. Thank you for your time!

[SiddharthSingi]

Someone please approve this PR, its causing all basic metrics to fail. Even the basic examples given on the evaluate metric websites are not working

[SiddharthSingi]

This checks for the type of the output and solves the error of float not having a size error.

[tomaarsen]

Thanks for raising this! I'm afraid evaluate is no longer actively being maintained, but this is such a small fix that I figured I'd merge it.
I believe the commit should over time automatically be propagated to the Hugging Face spaces, after which the problem should be resolved fully.

• Tom Aarsen