o 8wi5 @slddlZddlZddlmZddlmZddlmZddl m Z ddl m Z ddl m Z Gd d d eZ dS) N)Optional)Pipeline)Uniform)cdist) dist_range) l2_normalizecs>eZdZdZ d deedeeffdd Zdd ZZ S) ClosestAssignmentagAssign each sample to the closest target Parameters ---------- metric : `str`, optional Distance metric. Defaults to 'cosine' normalize : `bool`, optional L2 normalize vectors before clustering. Hyper-parameters ---------------- threshold : `float` Do not assign if distance greater than `threshold`. cosineFmetric normalizecs`t||_||_t|j|jd\}}t|s(d}d|dd}t|t |||_ dS)N)r r g.Azbounding distance threshold to gz(: this might lead to suboptimal results.) super__init__r r rnpisfinitewarningswarnr threshold)selfr r min_distmax_distmsg __class__d/home/ubuntu/sommelier/.venv/lib/python3.10/site-packages/pyannote/pipeline/blocks/classification.pyr8s    zClosestAssignment.__init__cCsf|jr t|}t|}t|||jd}tj|dd}t|D]\}}|||f|jkr0| ||<q|S)aAssign each sample to its closest class (if close enough) Parameters ---------- X_target : `np.ndarray` (n_targets, n_dimensions) target embeddings X : `np.ndarray` (n_samples, n_dimensions) sample embeddings Returns ------- assignments : `np.ndarray` (n_samples, ) sample assignments )r r)axis)r rrr rargmin enumerater)rX_targetXdistancetargetsikrrr__call__Ks zClosestAssignment.__call__)r F) __name__ __module__ __qualname____doc__rstrboolrr& __classcell__rrrrr (sr )rnumpyrtypingrpipeliner parameterrpyannote.core.utils.distancerrrr rrrrs