§
    à©ºjý  ã                  óŒ   — d Z ddlmZ ddlmZ ej        Zej        Zej        Zej	        Z	ej
        Z
ej        Zej        Zg d¢ZdS )u»  Metric functions â€” numbers computed from matches.

Every function here takes plain arrays and returns plain values. No evaluator, no
dataset, no COCO JSON::

    from hotcoco import metrics

    ap = metrics.average_precision(scores, matched, num_gt=len(ground_truths))
    ece, mce = metrics.calibration_error(scores, matched)

This is the shape ``sklearn.metrics`` and ``torchmetrics.functional`` use, and for
the same reason: a metric is a pure function of its inputs, so tying it to an
evaluator object only makes it harder to reach. ``COCOeval`` is still there when
you want the whole COCO pipeline â€” its analysis methods call straight into these.

``metrics`` produces numbers from matches; :mod:`hotcoco.primitives` produces the
matches themselves. ``primitives.lsap`` decides *which* prediction pairs with which
ground truth; ``metrics.average_precision`` turns that decision into a number.

Stability
---------

Provisional through the 1.x series â€” public so drivers and users can share them,
but not frozen until 1.4. Expect additive change rather than reshaping of what is
here. ``COCOeval`` and the ``pycocotools`` drop-in surface are the permanent,
frozen part of the API and are unaffected.
é    )Úannotationsé   )Úmetrics)Úaverage_precisionÚcalibration_curveÚcalibration_errorÚconfusion_matrixÚis_computedÚ
is_missingÚprecision_recall_curveN)Ú__doc__Ú
__future__r   Úhotcocor   Ú_metricsr   r   r   r   r	   r
   r   Ú__all__© ó    ú_/root/private/project/pallet_detection_v2/.venv/lib/python3.11/site-packages/hotcoco/metrics.pyú<module>r      s‡   ððð ð8 #Ð "Ð "Ð "Ð "Ð "à (Ð (Ð (Ð (Ð (Ð (àÔ.Ð Ø!Ô8Ð ØÔ.Ð ØÔ.Ð ØÔ,Ð ØÔ"€ØÔ €
ðð ð €€€r   